Fast Apriori Algorithm for Frequent Itemset Mining

    

S. C. Mehrotra Ratnadeep R. Deshmukh Sachin N. Deshmukh C. Namrata Mahender Pravin L. Yannawar Department of Computer Science & Information Technology Dr. Babasaheb Ambedkar Marathwada University Aurangabad, Maharashtra, India. Copyright © 2016, Editors and Excel Academy Publishers EXCEL ACADEMY PUBLISHERS

Gokulwadi, Aurangpura, Aurangabad-431001 (M. S) [email protected] All rights reserved. No part of this publication may be reproduced, stored in retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the publisher. All export rights for this book vest exclusively with Editors and Excel Academy Publishers. Unauthorized export is a violation of terms of sales and is subject to legal action. ISBN 978-93-86751-04-1 Published by Excel Academy Publishers. Gokulwadi, Aurangpura, Aurangabad-431001 (M. S) Printed in India

Preface Cognitive Knowledge Engineering (CKE) is concerned with the application of computer systems to problems of human endeavor such as thinking, learning, problem solving, decision making, and knowledge transfer. At present, it refers to the building, maintaining and development of knowledge based systems. With the dramatic advances in data acquisition and storage technologies, the problem of how to turn raw data into useful information has become one of the most daunting problems of Cognitive Knowledge Engineering. Hence Cognitive knowledge engineering (CKE) and Data Mining are areas of common interest to researchers in AL, Pattern recognition, statistics, databases, knowledge acquisition, data visualisation, high performance computing, and expert systems. This volume brings together researchers, Scientist, Software developers and Vendors who have highlighted pioneering advance in the area of Cognitive Knowledge Engineering and its application to discuss and present their work. The chapters in this book were contributed at the IJCA International Conference on Cognitive Knowledge Engineering (ICKE -2016) organized by the Department of Computer Science and Information Technology of the Dr. Babasaheb Ambedkar Marathwada University, Aurangabad (Ms) India. The conference received an overwhelming response and after peer review, ninety nine papers were selected for presentation. The key areas on which the papers were received are Big Data Analytics, Human Computer Interaction, Biometric: Multimodal System Development, Natural Language Processing, Remote Sensing and GIS, Smart City and Smart Village, Signal Processing and Computer Vision. The contents of book are segregated in following six sections. Section 1: Sentiment analysis, use of Fast Apriori Algorithm, Certificateless Encryption, Context Based methods in Big Data Analytics section. Section 2: Soft computing, Genetic algorithm, Gesture Recognition in Human Computer Interaction section Section 3: Biometric applications, Person Identification, Disease Detection using Biomedical images like MRI, Image Processing Techniques in Biometric: Multimodal System Development section. Section 4: User Profile Modelling, role of Social Media in Natural Language Processing section. Section 5: Hyperspectral Remote sensing for various application area, Land Use Land Cover Mapping, Change Detection using Satellite Data, Solid Waste Management, Smart Phone Based system for agriculture in Remote Sensing and GIS, Smart City and Smart Village section. Section 6: Speech signal, Brain signal, Speaker recognition in Signal Processing and Computer Vision section. We are confident that this volume will be useful for researchers, industry and academics to augment their knowledge on the topic. We would like to acknowledge and thanks the contribution of UGC, SAP , DST-SERB, ISRO, ICMR, UGC-SAP, IETE, CSI, Endress Hauser, MIIT (Beed), CSMSS, in making this conference a success. We would like to thank all local organizations and Institutions for their sponsorship and support. The effort and contribution made by various officials and member of our parent organization Dr. Babasaheb Ambedkar Marathwada University, Aurangabad (Ms) India deserves special mention

vi Preface

and appreciation. We express our gratitude to all committee members, reviewers for spending their valuable time. We are also thankful to Padmashri Vijay Bhatkar, Chancellor of India International Multiversity, Chairman of ETH Research Lab, Chief Mentor of I2IT, and National President of Vijnan Bharati.India, Prof. Dr. Y. A. Kawade Mentor, MIT, and a Group of Academic & Research Institution Aurangabad (MS) India. Padmakar Muley, Secretory, CSMSS, Aurangabad (MS) India. ‘”ƒŽ •—’’‘”– ‘ˆ ‘—” ‘ǯ„Ž‡ ‹…‡-Chancellor Dr. B. A. Chopde and University Authorities has indeed added zest to our zeal in organizing this international conference. We take this opportunity to convey our sincere thanks to all the ICKE-2016 committee members for completing one or more inspiring event in a team spirit. We are also thankful to all the Teaching and Non-Teaching staff of the Department for their timely and constant support. We extend our thanks to all those who have directly or indirectly supported us in making this event a success. S. C. Mehrotra Ratnadeep R. Deshmukh Sachin N. Deshmukh C. Namrata Mahender Pravin L. Yannawar

Contents Preface  





















v



BigDataAnalytics  ͳǤ ’”‘˜‡‡–‹ ˆ‘”‡”‰›ˆˆ‹…‹‡–‹”‡Ž‡••‡•‘”‡–™‘”•

Ͳ͵

ƒ‹•ŠƒŽ‹ƒ”‘†‡ǡǤǤƒŠƒ‰ƒ–ǡƒ…Š‹ƒ”Šƒ–‡ 

ʹǤ ’Ž‡‡–ƒ–‹‘‘ˆ‹˜‹•‹‘ƒ†‡’Ž‹…ƒ–‹‘‘ˆƒ–ƒ‹Ž‘—†

Ͳ͹

ǤǤƒ–‹ǡǤǤƒƒ”™ƒŽ 

͵Ǥ ƒ•–’”‹‘”‹Ž‰‘”‹–Šˆ‘” ”‡“—‡– –‡•‡–‹‹‰

ͳ͵

ƒ–ƒ†‡‡’Ǥ‡•Š—ŠǡŠ—„Šƒ‰‹Ǥƒ–‹ŽǡŒƒ›‡•Š—Š 

ͶǤ ‡–‡…‡‡˜‡Ž‡‰ƒ–‹‘ †‡–‹ˆ‹…ƒ–‹‘ˆ”‘‡™•”–‹…Ž‡•—•‹‰‡–‹‡– ƒŽ›•‹•

ͳͻ

‹•ŠƒŽǤŠ‹”•ƒ–ǡƒŒ—ƒ”Ǥ ƒ‰†ƒŽ‡ǡǤǤ‡•Š—Š

 ͷǤ ƒŽ›œ‹‰–—†‡–•‡”ˆ‘”ƒ…‡•‹‰…ƒ†‡‹…ƒŽ›–‹…•

ʹʹ

‹‡–—Ž‡›ǡƒ”ƒ‰ŠƒŽ…Šƒ†”ƒǡƒŠ‡•Š ‘•Š‹ 

͸Ǥ ‘˜‡Ž’’”‘ƒ…Šˆ‘”‹’”‘˜‹‰‡…—”‹–›ƒ†‘ˆ‹†‡–‹ƒŽ‹–›‹—„Ž‹…Ž‘—†• •‹‰‡”–‹ˆ‹…ƒ–‡Ž‡••…”›’–‹‘

ʹ͸

Š—†ƒƒ‡˜‹†ƒ•”ƒ‘—–‡ǡƒ—‡ŽǤǡƒ–ƒ†‡‡’Ǥ‡•Š—Š 

͹Ǥ ‘ˆ–™ƒ”‡”‹‡–‡†‹•Ž‡••Ž‹‡–ˆ‘”‘•–ˆˆ‹…‹‡…›

͵ʹ

ǤǤ‘ƒ™ƒ‡ǡǤǤ‡•Š—ŠǡǤǤƒ‰Šƒ”‡ǡ—•Š’‡†”ƒŠƒ˜ƒ 

ͺǤ ‘’ƒ”ƒ–‹˜‡–—†›‘ˆ‡–‹‡–ƒŽ›•‹•’’”‘ƒ…Š‡•‘†‹ˆˆ‡”‡–†ƒ–ƒ•‡– •‹‰ƒ…Š‹‡Ž‡ƒ”‹‰‡…Š‹“—‡•

͵͸

•Š‹•ŠŠƒŽ‡”ƒ‘ǡƒ…Š‹‡•Š—Š 

ͻǤ –—†›‘ˆƒ”‹‘—•’’”‘ƒ…Š‡•‘ˆ‡ƒ•Ž‰‘”‹–Š–‘ ’”‘˜‡ˆˆ‹…‹‡…›

ͶͲ

‹Žƒ•ǤƒŠƒ–‡ǡŠ‹–ƒŽǤƒ†‡ 

ͳͲǤ ‘–‡š–ƒ•‡†‡…‘‡†ƒ–‹‘‡–Š‘†•ǣ”‹‡ˆ‡˜‹‡™

Ͷ͵

”ƒ–‹Ǥ‡•Š’ƒ†‡ǡƒ—‡ŽǤ 

ͳͳǤ —–‘ƒ–‹… †‡–‹ˆ‹…ƒ–‹‘‘ˆ”‘•ƒ†‘•‘ˆ”‘†—…–ˆ”‘™‹––‡”

ͷͲ

‹– ǤŠ‹”„Šƒ–‡ǡ‡ŠƒǤƒƒ†‡

 ͳʹǤ —”˜‡›‘‘’‹…‘†‡Ž‹‰ˆ‘” ‡ƒ–—”‡Ǧƒ•‡†’‹‹‘‹‹‰

ͷͷ

ƒ†ƒ’ƒ‹Ǥ”‹„Š—˜ƒǡ—‹Ž ǤŠ‹”—†ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š

 ͳ͵Ǥ ‡˜‹‡™ˆ‘”’‹‹‘š–”ƒ…–‹‘ƒ†ƒŽ›•‹•„ƒ•‡†‘‡‹Ǧ—’‡”˜‹•‡† ‡ƒ”‹‰—•‹‰‘™Ž‡†‰‡ƒ•‡

͸Ͳ

ǤǤƒ†‡†ƒ”ǡ ǤǤƒ–‹ŽǡǤǤ ‘•Š‹ 

ͳͶǤ ‹‰ƒ–ƒƒŽ›•‹•‘ˆ†—…ƒ–‹‘––ƒ‹‡–‹ƒŠƒ”ƒ•Š–”ƒ–ƒ–‡

͸͵

ǤǤƒ–‹Žǡ’”ƒƒ•ŠǤ ƒ†Šƒ˜ 

ͳͷǤ ‹‰ƒ–ƒƒŽ›–‹…•ƒ†‡…—”‹–›‡ƒ•—”‡•

͸ͺ

”ƒ•Šƒ–ǤŠ‹–ƒŽǡƒ–ƒ†‡‡’Ǥ‡•Š—ŠǡƒŒ‡†”ƒ Ǥ ƒ‹™ƒ†

ͳ͸Ǥ ‹š‡†’’”‘ƒ…Šˆ‘”ƒ–ƒ‘›‹œƒ–‹‘—•‹‰ƒ’‡†—…‡ ”ƒ‡™‘”‘ Ž‘—†

͹Ͷ

ǤǤ ƒ†Šƒ˜ǡǤǤƒƒ”™ƒŽǡǤǤŠ‹†‡ 

ͳ͹Ǥ Žƒ••‹ˆ‹…ƒ–‹‘‘ˆ‡ƒ”–ƒ–ƒ—•‹‰

͹ͻ

ǤǤ—„Šƒ”ǡǤǤœƒǡǤǤ—†Š‘Žƒ” 

ͳͺǤ ”‡†•‘ˆ‘ƒ†ƒŽƒ…‹‰‹Ž‘—†…‡ƒ”‹‘ǣ‡˜‹‡™

ͺʹ

ƒ‡•ŠǤ —Œƒ”ǡƒ–‹•ŠǤ‡˜ƒ‡ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š 

ͳͻǤ ‡–Š‘†‘Ž‘‰›ˆ‘”ƒ–ƒƒƒ‰‡‡–‹—Ž–‹†‹‡•‹‘ƒŽƒ”‡Š‘—•‡

ͺͺ

‡”ƒ —ŽŠƒ–‹ƒŒ—Žƒǡ—ƒ ƒ›ƒ–ƒ‹‰Š

 ʹͲǤ ‡Ž‡…–‹‘‘ˆ‹†‘™‹œ‡ˆ‘” †‡–‹ˆ‹…ƒ–‹‘‘ˆ‡Ǧ—’Ž‹…ƒ–‡•™‹–Š—’Ž‹…ƒ–‡ ‘—––”ƒ–‡‰›

ͻ͸

ƒ‹•ŠƒŽ‹ƒ‰‹ƒ”ǡƒ…Š‹‡•Š—Šǡ—‹ŽŠ‹”—†

 HumanComputerInteraction  ʹͳǤ —”˜‡›‘‹•‹‘„ƒ•‡†•–ƒ–‹…ƒ††›ƒ‹…Šƒ†‰‡•–—”‡•›•–‡•ˆ‘” ‘„‹Ž‡†‡˜‹…‡•

ͳͲͳ

ǤǤ‡Šƒƒ”ǡƒŒ‡‡˜ ƒ‹ǡǤǤŠƒƒ”‡ 

ʹʹǤ ‡˜‹‡™‘ˆ‘ˆ–™ƒ”‡ˆˆ‘”–ƒ†‘•–•–‹ƒ–‹‘—•‹‰‘ˆ–‘’—–‹‰

ͳͲ͸

œƒ†‡ƒŒƒ›Ǥǡ‘–‡ǤǤǡ“—‡‡ŽŠ‡†Ǥ ƒŽ‹Ž 

ʹ͵Ǥ ‘Ž˜‹‰•‹‰ ‡‡–‹…Ž‰‘”‹–Š’’”‘ƒ…Š

ͳͳͲ

ƒ”‡‡‡Ž‹ǡǤǤƒƒ”™ƒŽ 

ʹͶǤ ‡Šƒ˜‹‘”‘ˆŽ‡…–”‘ƒ‰‡–‹…ƒ˜‡•ƒ––‡ƒ‘—†ƒ”›‹‹”‡Ž‡•• ‘—‹…ƒ–‹‘›•–‡ǣ–—–‘”‹ƒŽ

ͳͳͷ

Šƒ‘Š‡Ž”ƒƒǡƒ››ƒ†Œ‹ŒǤ 

ʹͷǤ —”˜‡›‘‹•‹‘Ǧ„ƒ•‡†ƒ† ‡•–—”‡‡…‘‰‹–‹‘

ͳʹʹ

—‹Ž Ǥ‡•Š—ŠǡǤǤ ƒ‰ƒ†‡

 ʹ͸Ǥ ’–‹‹œ‹‰–Š‡‡•‘”‡“—‹”‡‡–ˆ‘” ‡•–—”‡‡…‘‰‹œ‹‰›•–‡•„› ‹†‹‰–Š‡‘•–‹‰‹ˆ‹…ƒ–‡•‘”••‹‰–ƒ–‹•–‹…ƒŽ‡–Š‘†•

ͳʹͺ

†”‡™•ƒ”ƒŒǡ”ƒ„ƒƒ”ƒǤǡƒ–Š‹•ŠǤ 

ʹ͹Ǥ  Ǧ ‡Ǧ‡ƒ”‹‰‘”–ƒŽǣ‘†‡Ž‘ˆ ˆ‘”†—…ƒ–‹‘ƒŽ‡˜‡Ž‘’‡–

ͳ͵͵

ǤǤƒ”‡ǡ ǤǤ ƒ˜ƒŽ‡ǡǤƒ™ƒ”ǡ ǤǤŠƒ‹ŠǡǤǤƒ”‡ 

ʹͺǤ —Ž–‹Ž‡˜‡Ž‘—…ŠȀ”‡••‡…‡’–‘”Ǧƒ•‡†Š‡”ƒ’‡—–‹…š‡”…‹•‡ˆ‘” –—†‡–•™‹–Š’‡…‹ƒŽ‡‡†•

ͳ͵ͺ

†”‡™•ƒ”ƒŒǡ”ƒ„ƒŠƒ”ƒǤǡƒŽ˜‹ƒƒŒ‡†”ƒ

 ʹͻǤ ˆˆ‡…–‘ˆ‡’–Šƒ”‹ƒ–‹‘‘–Š‡‹‰Ǧ•Šƒ’‡†‹…”‘‹š‡”

ͳͶʹ

”•Š‹›ƒŒ—ŠƒǡǤǤƒ†‘ǡ ‹”†‘—•ŠƒǡǤǤ‘Ž‡ 

͵ͲǤ ‡ƒ–—”‡”ƒ…‹‰—•‹‰ƒ”–‹…Ž‡ ‹Ž–‡”‹‘’‡‹’’‹‰ˆ‘” ”‘••‘–‘”‹ŽŽ ‡˜‡Ž‘’‡–

ͳͶ͸

–‡’Š‡ƒ”—‰ƒ”—ǡ‡Œ‹ƒ–•——”ƒǡƒ†ƒ ‘–‘†ƒ 

Biometric:MultimodalSystemDevelopment  

͵ͳǤ ͶǤͷƒ†ˆ‘”–”—…–—”ƒŽ †‡–‹ˆ‹…ƒ–‹‘‘ˆ”›•–ƒŽŽ‹‡‘Ž‹†•ǣ‡˜‹‡™ ƒ˜‹†”ƒǤƒ–‹ŽǡǤǤƒƒ”™ƒŽǡǤǤ—”—†‡



ͳͷ͵

͵ʹǤ ‡”ˆ‘”ƒ…‡˜ƒŽ—ƒ–‹‘‘Ǧ ƒ…‡ƒ–ƒ„ƒ•‡—•‹‰—Ž–‹Ž‰‘”‹–Š‹… —Ž–‹‡•‘”’’”‘ƒ…Š

ͳͷ͸

‹††Šƒ”–ŠǤƒ„Šƒ†‡ǡƒ‰•‡Ǥƒ•‘†ǡǤǤƒŽ‡ǡǤǤƒœ‹ǡ‘‰‡•ŠǤ‘†‡ 

͵͵Ǥ ƒŽ’”‹–’‡…–”ƒŽ‹‘‡–”‹…—–Š‡–‹…ƒ–‹‘›•–‡„›—•‹‰ ‹‡Ž†’‡…Ͷ’‡…–”‘”ƒ†‹‘‡–‡”

ͳ͸͵

‹–ƒ ǤŠƒ†‹œ‘†ǡƒ–†‡‡’Ǥ‡•Š—Šǡƒ†ƒƒǤ ƒ†Šƒ˜ƒ–‹Ž 

͵ͶǤ ‘˜‡Ž’’”‘ƒ…Šˆ‘”‡”•‘ †‡–‹ˆ‹…ƒ–‹‘ƒ•‡†‘ ”‹•–ƒ–‹•–‹…ƒŽ ‡ƒ–—”‡• ƒ†‡–‹ƒŽŽ‘‘†‡••‡Ž•‹ˆ—”…ƒ–‹‘‘‹–•

ͳ͸ͺ

‘‰‡•ŠƒŒ’—–ǡƒ‡•Šƒœƒǡ—’”‹›ƒƒ„Ž‡ǡ—•Š‹Ž ƒ™ŠƒŽ‡ǡ„†—Žƒƒ„†—Ž ƒƒ 

͵ͷǤ —”˜‡›‘ƒ–ƒ‹†‹‰‡…Š‹“—‡•

ͳ͹Ͷ

ƒ‹•ŠƒŽ‹Ǥ ‰Ž‡ǡǤǤƒ–‹ŽǡǤǤŠƒ‰ƒ–ǡ ǤǤŠƒ—†Šƒ”‹ 

͵͸Ǥ ƒŽ›•‹•‘ˆ”‘•–ƒ–‡ƒ…‡”‹ —•‹‰—’’‘”–‡…–‘”Žƒ••‹ˆ‹‡”•ƒ† ‘‹•–ƒ…‡„ƒ•‡†Ž—•–‡”‹‰

ͳͺͳ

‹–‡•Š‹Ž—ƒ” ƒ‰™ƒŽǡƒ–ƒ†‡‡’Ǥ‡•Š—Šǡƒ—‡ŽǤ

 ͵͹Ǥ ‡ƒˆ‡ƒ–‹‘š–”ƒ…–‹‘ˆ”‘‡ƒˆ ƒ‰‡•ǣ”‡˜‹‡™

ͳͺͷ

”ƒ†‹’ƒŽ˜‡ǡ‹Ž‹†ƒ”†‡•ƒ‹ǡ”ƒ˜‹ƒƒ™ƒ”

 ͵ͺǤ ‘…‡’–—ƒŽ‘†‡Žˆ‘”—–‘ƒ–‡†––‡†ƒ…‡›•–‡•‹‰”‹…‹’ƒŽ ‘’‘‡–ƒŽ›•‹•ȋȌ

ͳͻͲ

ƒ”‹ƒ•Š‘‘˜‹–ƒ”ǡ‡‡ƒǤƒ™ƒ–Š‡ƒ”

 ͵ͻǤ ‘’ƒ”ƒ–‹˜‡ƒƒŽ›•‹•‘ˆ‡†‰‡†‡–‡…–‹‘–‡…Š‹“—‡•ˆ‘”ƒƒŽ›œ‡‹ƒ„‡–‹… ‡’Š”‘’ƒ–Š›—•‹‰”‡ƒŽ‹‘’•‹‡•‹ƒ‰‡•

ͳͻͶ

‘‰‹‹Ǥƒ–‹Žǡ‡‡ƒƒ˜ƒ–‡ƒ” 

ͶͲǤ ‡˜‹‡™‘ ƒ‰‡’”‘…‡••‹‰‡…Š‹“—‡•ˆ‘” Žƒ—…‘ƒ‡–‡…–‹‘

ͳͻͺ

ƒ‰ƒ†‡˜‹Ǥ‡†‡ǡ—–‹ ƒ†Šƒ˜ǡƒ‡•ŠǤƒœƒǡ™ƒ’ƒŽ‹ƒŽŽƒ†‡ 

ͶͳǤ ”‡†‹…–‹‘‘ˆ—ƒ…–‹˜‹–›‹‹†‡‘

ʹͲ͵

•Š™‹‹Ǥ ƒ˜ƒŽ‹ǡǤǤƒƒ”™ƒŽǡǤǤŠ‹†‡ 

ͶʹǤ ‘’—–‡”••‹•–‡†ƒŽ›•‹•ƒ†›•–‡‹œƒ–‹‘‘ˆ‡‡•–‡‘ƒ”–Š”‹–‹•—•‹‰ ‹‰‹–ƒŽǦ”ƒ› ƒ‰‡•

ʹͲ͹

Š‹˜ƒƒ†Ǥ ‘”ƒŽ‡ǡ‘‘ŒƒǤƒ–”ƒ˜ƒŽ‹ǡƒ‡•ŠǤƒœƒ 

Ͷ͵Ǥ ••‘…‹ƒ–‹‘‡–‡…–‹‘‘ˆ‡‰—Žƒ” •—Ž‹ƒ† •—Ž‹•‹‰–ƒ–‹•–‹…ƒŽ ‡ƒ–—”‡•

ʹͳ͵

—’”‹›ƒǤƒ„Ž‡ǡŠƒ‹Š„†—Žƒƒǡ‘‰‡•ŠǤƒŒ’—–ǡ›ƒ‡•Š™ƒ”‹Ǥƒ–‹Ž

 ͶͶǤ ‘‡‡”‰›ƒƒ‰‡‡–›•–‡™‹–Š‡‡™ƒ„Ž‡‡”‰›‘—”…‡•ƒ•‡†  

ʹͳ͹

Š‡–ƒǤƒ†‡ǡ ƒ‡•ŠǤƒ„Ž‡

 ͶͷǤ ‡˜‹‡™ǣ‹ˆ‹‡† ’’”‘ƒ…Š–‘‹•—ƒŽ’‡‡…Šƒ†’‡ƒ‡”‡…‘‰‹–‹‘

ʹͳͻ

‹–‡•ŠǤƒ‰”‡ǡŒ‹–Ǥ Š‘†‡

 Ͷ͸Ǥ ‡˜‹‡™Ȃ‘’—–‡”ƒ•‡†‹’‡ƒ†‹‰›•–‡ˆ‘”ƒ”ƒ–Š‹ƒ‰—ƒ‰‡

ʹʹ͵

Š—•ŠƒǤ—Žƒ”‹ǡŒ‹–Ǥ Š‘†‡ 

Ͷ͹Ǥ ƒ–ƒ‡…‘˜‡”ˆ”‘‡…‘†ƒ”›–‘”ƒ‰‡‡˜‹…‡ˆ–‡”—Ž–‹’Ž‡‹‡•‡Ž‡–‹‰ ‘”ƒ––‹‰ƒ†‹’‹‰ ƒ‹•ŠƒǤŠ‘’•‡ǡŠƒ”ƒ•‹‰Ǥƒ›–‡ 

ʹʹ͹

ͶͺǤ ƒ‰‡ …‘•‹•–‡…›‡–‡…–‹‘ƒ•‡†‘ ”ƒ†‹‡–‹”‡…–‹‘

ʹ͵ͳ

ƒŠƒŽ‡‹˜‡‹ŽƒŽǡ”ƒ˜‹ƒƒ™ƒ”ǡ•Š‘Ǥ ƒ‹™ƒ†

 ͶͻǤ ‡ƒ–—”‡‡˜‡Ž —•‹‘ˆ‘” ‹‰‡”’”‹–—•‹‰‡—”ƒŽ‡–™‘”ˆ‘”‡”•‘ †‡–‹ˆ‹…ƒ–‹‘

ʹ͵Ͷ

‹††‹“—‹Žƒ•ǡ‘–Š‡ƒ˜‹–ƒǤǡ‡Ž‰ƒ†—’ƒŽ‹Ǥǡ‡•Š—ŠǤǤ

 ͷͲǤ ‹‘‡–”‹…”›’–‘•›•–‡ˆ‘” ‹‰‡”’”‹–‡”‹ˆ‹…ƒ–‹‘

ʹͶͲ

Š—„Šƒ‰‹ƒ’ƒŽǡƒ–ƒ†‡‡’‡•Š—Š

 ͷͳǤ ‡ƒ–—”‡š–”ƒ…–‹‘‘ˆ ”‹•ƒ•‡†‡–‡…–‹‘‘ˆ‹‰Š‡”‰›‹†‰‡•

ʹͶͷ

‹–ƒǤƒ–‹Žǡ‹”—’ƒƒǤƒ–‘†ƒ”ǡ”ƒ’–‹Ǥ‡•Š—Š

 ͷʹǤ ‡ƒˆ˜‡ƒ–‹‘‡š–”ƒ…–‹‘„›‘”’Š‘Ž‘‰‹…ƒŽ‘’‡”ƒ–‹‘ˆ‘”Žƒ– Žƒ••‹ˆ‹…ƒ–‹‘

ʹͶͻ

ƒ‹•ŠƒǤŽ‡ƒ”ǡ•Š‘Ǥ ƒ‹™ƒ†

 ͷ͵Ǥ ”‡ƒ•–ƒ••‡‰‡–ƒ–‹‘—•‹‰‡‡†ƒ•‡†‡‰‹‘ ”‘™‹‰‡…Š‹“—‡

ʹͷʹ

‘–Š‡ƒ˜‹–ƒǤǡ‡Ž‰ƒ†—’ƒŽ‹Ǥǡ‹††‹“—‹Žƒ•ǡ‡•Š—ŠǤǤ

 ͷͶǤ ‡˜‹‡™‘—Ž–‹‘†ƒŽ‹‘‡–”‹…—ƒ‡…‘‰‹–‹‘›•–‡—•‹‰͵ ƒ…‡ƒ†͵ƒ”

ʹͷ͸

—‡‰ŠŠƒ”‡™ƒŽǡǤǤƒŽ‡

 ͷͷǤ ‡˜‹‡™ǣƒŽ’”‹–‡…‘‰‹–‹‘”‘…‡••ƒ†‡…Š‹“—‡•

ʹ͸ʹ

‘—ƒ†ǤǤŽ‹ǡ”ƒ˜‹ƒƒ™ƒ”ǡǤǤ ƒ‹™ƒ†

 ͷ͸Ǥ —‡Žˆˆ‹…‹‡–”‹˜‹‰‡Šƒ˜‹‘”•‹‰š’‡”–›•–‡•ǣ‡˜‹‡™

ʹ͸ͻ

ƒŒ—ƒ–ŠǤǤǡǤƒŠ‡•Š™ƒ”‹ǡŠƒ”‹ŽƒŠ‹†ƒ”ƒ˜ƒŽŽ‹ǡ†”‡™•ƒ”ƒŒ

 ͷ͹Ǥ –—†›‘‹ƒ‰‘•‹‰ ‡ƒ”„‘š ƒ—Ž–•‹‡Š‹…Ž‡••‹‰š’‡”–›•–‡•ǣ ‡˜‹‡™

ʹ͹ͷ

Šƒ”‹ŽƒŠ‹†ƒ”ƒ˜ƒŽŽ‹ǡ ‡‹Žƒ‹˜‹‰•–‘ǤǤǡƒŒ—ƒ–ŠǤǤǡ†”‡™•ƒ”ƒŒ

 ͷͺǤ ‹‰‡”’”‹–ƒŽ›•‹•„›ˆ—•‹‘‘ˆ‹‰—Žƒ”ƒŽ—‡‡…‘’‘•‹–‹‘ƒ† ‹•…”‡–‡ƒ˜‡Ž‡–”ƒ•ˆ‘”ˆ‘”‰‡†‡”‹†‡–‹ˆ‹…ƒ–‹‘

ʹͺͳ

ƒ‡•ŠǤ‘‰”‡ǡǤǤ ƒ‰ƒ†‡

 ͷͻǤ ƒŽ›•‹•‘ˆˆ‡ƒ–—”‡”ƒ‹‰ƒ†ˆ‡ƒ–—”‡•—„•‡–•‡Ž‡…–‹‘ƒŽ‰‘”‹–Šˆ‘” ’‹Ž‡’–‹…‡‹œ—”‡”‡†‹…–‹‘

ʹͺ͹

Œ—Šƒ‹ŠǡŠ—„Šƒ•Š”‡‡ƒ˜ƒ–ǡ—–ƒŠ‘’‡•Š™ƒ”ƒ”

 ͸ͲǤ ˆˆ‡…–‘ˆ‹‹Žƒ”‹–›‡ƒ•—”‡•ƒ††‡”Ž›‹‰—„•’ƒ…‡‘‹‡ƒ” ‹•…”‹‹ƒ–ƒŽ›•‹•

ʹͻͳ

—•Šƒ‹‡–‘”ƒ†‡ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š 

͸ͳǤ ‡•‹‰‘ˆ‡—”‘ —œœ››•–‡ˆ‘”†‹ƒ‰‘•‹•‘ˆ”‡†‹ƒ„‡–‡•ƒ†›’‡Ǧʹ ‹ƒ„‡–‡•

ʹͻͶ

ǤǤ„‹Ž™ƒ†‡ǡǤǤƒœƒǡƒ‹•ŠƒŽ‡ƒ”

 ͸ʹǤ ‘’ƒ”‹•‘‘ˆŠƒ…‡‡–‡…Š‹“—‡•ˆ‘”–Š‡‡––‡”‡–‘ˆ‡–ƒŽ ƒ†‹‘‰”ƒ’Š ‹”—’ƒƒǤƒ–‘†ƒ”ǡ‹–ƒǤƒ–‹Žǡ”ƒ’–‹Ǥ‡•Š—Š  

 

ʹͻͺ

NaturalLanguageProcessing  ͸͵Ǥ ˜ƒŽ—ƒ–‹‘‘ˆ‡‡˜ƒŽʹͲͳ͸‡•–ƒ—”ƒ–ƒ†ƒ’–‘’‡˜‹‡™•ƒ–ƒˆ‘” ‘–‹‘ƒ†•’‡…–Žƒ••‹ˆ‹…ƒ–‹‘

͵Ͳ͹

ƒ–ƒ†‡‡’Ǥ‡•Š—Šǡ›ƒ‡•Š™ƒ”Ǥ‹”ƒ‰‡ǡŒƒ›‡•Š—Š

 ͸ͶǤ —–‘ƒ–‹…‡š–Žƒ‰‹ƒ”‹•‡–‡…–‹‘ˆ”‘‡•‡ƒ”…Šƒ’‡”•‹‰‹‰ ‹˜‡ ‡”•‘ƒŽ‹–›”ƒ‹–•ˆ”‘™‹––‡”ƒ–ƒǣ‡˜‹‡™

͵ͳʹ

”—•ŠƒŽ‹Š—›ƒ”ǡƒ†Š—”‹ ‘•Š‹

 ͸ͷǤ ‹‹‰‘˜‹‡ –‡–‹‘•‹‰ƒ›‡•ƒ†ƒš‹—–”‘’›Žƒ••‹ˆ‹‡”•

͵ͳ͸

ƒ”•ŠƒǤ ƒ†Šƒ˜ǡƒ…Š‹Ǥ‡•Š—Š

 ͸͸Ǥ ‘…‡’–—ƒŽ‡’‡†‡…›ˆ‘”—†‡”•–ƒ†‹‰“—‡•–‹‘ǣ•›•–‡

͵ʹ͸

ƒŽ’ƒƒŠƒ†ƒŽ‡ǡƒ—ƒ–Ǥ ‹–‡ǡǤƒ”ƒ–ƒƒŠ‡†‡”

 ͸͹Ǥ —‡•–‹‘ƒ•‡†‡š–—ƒ”‹œƒ–‹‘—†‡”—Ž‡ƒ•‡† ”ƒ‡™‘”

͵͵ʹ

‡‡’ƒŽ‹Ǥ ƒ‹™ƒ†ǡƒ‹•ŠƒŽ‹ƒ†ƒǡǤƒ”ƒ–ƒƒŠ‡†‡”

 ͸ͺǤ ‘…‹ƒŽ‡†‹ƒ‘Ž‡‹”‘—‰Š–…ƒ”…‹–›™ƒ”‡‡••

͵͵͸

—”ƒ„Š‹ǤŠ‘”ƒ–ǡ”‹›ƒƒ ƒ‹™ƒ†ǡǤƒ”ƒ–ƒƒŠ‡†‡”

 ͸ͻǤ ‡˜‡Ž‘’‡–‘ˆƒ”ƒ–Š‹‡š–‘”’—•ˆ‘”Žƒ‰‹ƒ”‹•‡–‡…–‹‘‹ƒ”ƒ–Š‹ ƒ‰—ƒ‰‡

͵ͶͲ

ƒ‡•ŠǤƒ‹ǡƒŠ‡•Š—ƒ”Ǥƒ†‰‡ǡǤƒ”ƒ–ƒƒŠ‡†‡”

 ͹ͲǤ –‘”›—ƒ”‹œƒ–‹‘—•‹‰‡š–‘’ƒ…–‘”

͵Ͷͷ

‡‡’ƒŽ‹ƒŠƒŒƒǡ‡‡’ƒŽ‹ƒ™ƒ‡ǡǤƒ”ƒ–ƒƒŠ‡†ƒ”

 ͹ͳǤ –—†›‘ˆ•‡””‘ˆ‹Ž‡‘†‡Ž‹‰ˆ‘”‡”•‘ƒŽ‹œ‡†‡…‘‡†‡”›•–‡

͵Ͷͺ

ǤǤ‹ƒ”‹™ƒŽƒǡǤǤ‡•Š—Š

 ͹ʹǤ ‡…‘‡†‡”›•–‡•ƒ–ƒ Žƒ…‡

͵ͷʹ

Šƒ–‹Š‹”™ƒ†ƒ”ǡƒ…Š‹‡•Š—Š

  RemoteSensing&GIS,SmartCities&SmartVillage  ͹͵Ǥ ›’‡”•’‡…–ƒŽ‡‘–‡‡•‹‰ˆ‘”‰”‹…—Ž–—”‡ǣ‡˜‹‡™

͵ͷͻ

‘‘Œƒ‹‘† ƒ•‡ǡƒ–ƒ†‡‡’Ǥ‡•Š—Šǡ–‡’Š‡ Ǥƒ”—‰ƒ”—,ǤǤƒŽ›ƒƒ”ǡƒƒŒ‘‹ƒ” 

 ͹ͶǤ †˜ƒ…‡‘„‹Ž‹–›™ƒ”‡›„”‹†”‘ƒ†…ƒ•–‹‰‹

͵͸Ͷ

ƒŒ—•Šƒ‡•Š—Šǡƒ‰‡‡–ƒƒƒ”™ƒŽ 

͹ͷǤ ƒŽ›•‹•‘ˆˆˆ‡…–•‘ˆ‹”‘ŽŽ—–‹‘‘ƒ‰‘ƒ†—•–ƒ”†ƒ’’Ž‡”‡‡‡ƒ˜‡• •‹‰ ‹‡Ž†’‡…Ͷ’‡…–”‘”ƒ†‹‘‡–‡”ƒ†’‡…–”ƒŽ †‹…‡•

͵͸ͺ

”…ŠƒƒǤƒ–‡ǡ™ƒ–‹Ǥƒ‰ƒ”‡ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š

 ͹͸Ǥ ›’‡”•’‡…–”ƒŽ‡‘–‡‡•‹‰ˆ‘”–Š‡–—†›‘ˆŠƒ”ƒ…–‡”‹•–‹…•‘ˆ”‘’•

͵͹ͷ

ŒƒƒǤ Š—Ž‡ǡǤǤ‡•Š—Šǡ‘‘ŒƒǤ ƒ•‡

 ͹͹Ǥ ƒŽ›•‹•‘ˆˆˆ‡…–•‘ˆʹǡͶǦ‘†‹—ƒŽ–‘ŠŽ‘”‘’Š›ŽŽ‘–‡–‹ ‘™ƒ” —•‹‰ ‹‡Ž†’‡…Ͷ’‡…–”‘”ƒ†‹‘‡–‡” ™ƒ–‹Ǥƒ‰ƒ”‡ǡ”…ŠƒƒǤƒ–‡ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š



͵͹ͻ

͹ͺǤ ƒŽ›•‹•‘ˆŠŽ‘”‘’Š›ŽŽƒ†ƒ–‡”‘–‡–‘ˆ ƒ”†‡‹ƒ‡•‹‹ˆ‡”ƒ•‹‰‘Ǧ ‡•–”—…–‹˜‡‡…Š‹“—‡

͵ͺ͵

ƒŠ—ŽǤƒŠƒ”ƒ”ǡƒ–ƒ†‡‡’Ǥ‡•Š—Šǡƒ‹„Šƒ˜Ǥ‘Šƒ†‡

 ͹ͻǤ ƒ†•‡ƒ†‘˜‡”ƒ’’‹‰ƒ†Šƒ‰‡‡–‡…–‹‘„›•‹‰‡‘–‡ ‡•‹‰ Ǧ ‡’‘”ƒŽƒ–ƒ•‡–•‘ˆ—”ƒ‰ƒ„ƒ†‹–› 

͵ͺ͹

Œƒ›Ǥƒ‰‡ǡƒŒ‡•ŠǤŠ—ƒŽǡ‘ŽǤ‹„Š—–‡ǡǤǤƒŽ‡ǡǤǤ‡Š”‘–”ƒ

 ͺͲǤ ‡˜‹‡™‘ˆ‹‰‹–ƒŽ‘‹Žƒ’’‹‰”‘…‡†—”‡•

͵ͻͳ

ƒ›–‡ ƒ›’ƒŽ•‹‰ƒ––Š•—‹‰ǡǤǤ‡•Š—ŠǡǤǤƒŽ›ƒƒ”

 ͺͳǤ ”‡†‹…–‹‘‘ˆ”•‡‹…‘–‡–‹‘‹Ž—•‹‰‡ˆŽ‡…–ƒ…‡’‡…–”‘•…‘’›

͵ͻ͸

ƒ‹„Šƒ˜Ǥ‘Šƒ†‡ǡƒŠ—ŽǤƒŠƒ”ƒ”ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š

 ͺʹǤ ƒ–Š‡ƒ–‹…ƒŽ‘†‡Ž—•‡†‹‘Ž‹†ƒ•–‡ƒƒ‰‡‡–

͵ͻͻ

ǤǤŠ‹…Š‘†ƒ”ǡ’”ƒƒ•ŠǤ ƒ†Šƒ˜

 ͺ͵Ǥ ƒŽ›•‹•‘ˆ’Ž‘›‡–‡•‹–›ˆ‘”—•–ƒ‹ƒ„Ž‡”„ƒ‡‰‡‡”ƒ–‹‘•‹‰

‡‘Ǧ’ƒ–‹ƒŽ‡…Š‘Ž‘‰›

ͶͲͶ

ƒŠ—ŽǤƒ‰ƒ”ǡ ƒ‡•ŠǤƒ‰ƒ”ǡƒ–ƒ†‡‡’Ǥ‡•Š—Šǡƒ”„Šƒ”‹ǤƒŽ‡ǡ—”‡•ŠǤ‡Š”‘–”ƒ

 ͺͶǤ ‡ƒ–—”‡š–”ƒ…–‹‘„ƒ•‡†‘ƒ””‘™„ƒ†‡‰‡–ƒ–‹‘ †‹…‡•‘ˆ‘––‘ƒ† ƒ‹œ‡”‘’•„›Ǧͳ›’‡”‹‘ƒ–ƒ

ͶͲͻ

ƒŒ‡•ŠǤŠ—ƒŽǡŒƒ›Ǥƒ‰‡ǡ‘ŽǤ‹„Š—–‡ǡǤǤƒŽ‡ǡǤǤ‡Š”‘–”ƒ

 ͺͷǤ ‡–‡”‹ƒ–‹‘‘ˆ‡•–‹…‹†‡‹ƒƒƒ„› ˆ”ƒ”‡†’‡…–”‘•…‘’›•‹‰ƒ”–‹ƒŽ ‡ƒ•–“—ƒ”‡Ǧ‹•…”‹‹ƒ–ƒŽ›•‹•

Ͷͳ͵

‘Š‹‡‡Ǥ‹•ƒŽǡƒ–ƒ†‡‡’Ǥ‡•Š—Š

 ͺ͸Ǥ —Ž–‹Ǧ‡•‘”ǡ—Ž–‹Ǧ‡•‘Ž—–‹‘ƒ†—Ž–‹Ǧ‡’‘”ƒŽƒ–‡ŽŽ‹–‡ƒ–ƒ —•‹‘ˆ‘” ‘‹Ž›’‡Žƒ••‹ˆ‹…ƒ–‹‘

Ͷͳͺ

‘ŽǤ‹„Š—–‡ǡƒŒ‡•ŠŠ—ƒŽǡŒƒ›Ǥƒ‰‡ǡƒ†‡‡’ ƒ‹™ƒ†ǡƒ”„Šƒ”‹ǤƒŽ‡ǡǤǤ ‡Š”‘–”ƒ

 ͺ͹Ǥ ‡‘–‡•–‹ƒ–‹‘‘ˆŽƒ–•ŠŽ‘”‘’Š›ŽŽ‘–‡–•‹‰’‡…–”ƒŽ †‹…‡•–‘ †‡–‡…–…‹†ƒ‹

ͶʹͶ

ƒ…Š‹Ǥ‘ŽŠ‡ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š



ͺͺǤ ‘Ž†ƒ„Ž‡ƒ†……‹†‡–‡–‡…–‘”‡Ž‡–

Ͷʹͻ

›‘–‹—Žƒ”‡ǡ”ƒ–‹„Šƒƒ™ƒ”ǡƒŠ‡†”ƒ‡–Š‹

 ͺͻǤ ƒ”–Š‘‡„ƒ•‡†›•–‡ˆ‘”‰”‹…—Ž–—”‡”‘†—…–”‘…—”‡‡–ƒ† ‹•–”‹„—–‹‘

Ͷ͵͵

ǤǤƒ–‘•Š—ƒ”ǡǤƒƒ

 ͻͲǤ –ƒ†ƒ”†‡ƒ•—”‡‡–”‘–‘…‘Žˆ‘” ‹‡Ž†’‡…Ͷ’‡…–”‘”ƒ†‹‘‡–‡” —’ƒŽ‹Ǥ—”ƒ•‡ǡƒ”•‹ŠǤƒ”’‡ǡƒ”„Šƒ”‹ǤƒŽ‡         

 

Ͷ͵͹

SignalProcessing&ComputerVision  ͶͶ͵

ͻͳǤ ”ƒ‹‹‰ƒŽƒ–ƒˆ‘”‡…—”‹–›—”’‘•‡—•‹‰Ž‡…–”‘‡…‡’ŠƒŽ‘‰”ƒ’Š› ƒ‰•‡Ǥƒ•‘†ǡ‹††Šƒ”–ŠǤƒ„Šƒ†‡ǡǤǤƒŽ‡ǡǤǤƒœ‹ǡ‘‰‡•ŠǤ‘†‡

 ͻʹǤ ‡…‡–†˜ƒ…‡•‹‘Ž‘”„Œ‡…–‡…‘‰‹–‹‘ǣ‡˜‹‡™

ͶͶ͹

ƒŠ‡•ŠǤ‘Žƒƒ”ǡ”ƒ˜‹Ǥƒƒ™ƒ”

 ͻ͵Ǥ ‡š–Ǧ‡’‡†‡–—–‘ƒ–‹…’‡ƒ‡”‡…‘‰‹–‹‘›•–‡•‹‰‹†‹ƒ‰—ƒ‰‡

Ͷͷ͸

ƒ•Š‹Ǥ‰”ƒ™ƒŽǡǤǤ‡•Š—Šǡ™ƒ’‹ŽǤƒ‰Šƒ”‡

 ͻͶǤ ƒ”ƒ–Š‹ •‘Žƒ–‡†‹‰‹–‡…‘‰‹–‹‘›•–‡•‹‰

Ͷͷͻ

‡˜›ƒ‹Ǥ—Žƒ”‹ǡƒ–ƒ†‡‡’Ǥ‡•Š—Šǡƒ†ƒƒǤ ƒ†Šƒ˜ƒ–‹Žǡ™ƒ’‹ŽǤƒ‰Šƒ”‡ǡ —Š”ƒŒǤŠ”‹•Š”‹ƒŽǡƒ”‘Ǥ‹”‡”‡

 ͻͷǤ —–‘ƒ–‹…’‡‡…Š‡…‘‰‹–‹‘‡…Š‹“—‡•ǣ‡˜‹‡™

Ͷ͸Ͷ

„†—ŽƒŽ‹„†—ŽŽƒŠŽ‹Žƒ•ƒ†‹ǡƒ–ƒ†‡‡’Ǥ‡•Š—Š

 ͻ͸Ǥ —–‘ƒ–‹…‡…‘‰‹–‹‘‘ˆ–Š‡ƒ„Žƒ‘Ž••‹‰  ‡ƒ–—”‡

Ͷ͹ͳ

ƒ–‹•ŠǤƒƒ›‡ǡ—”‡•ŠǤ‡Š”‘–”ƒǡǤǤƒ†‘

 ͻ͹Ǥ —”˜‡›‘ Ž‘ƒ–‹‰‘‹–”‹–Š‡–‹…‹–•

Ͷ͹ͷ

Šƒ‹ŠǤǤǡ ‘†„‘Ž‡ǤǤ

 ͻͺǤ ‘‹…‡‡…‘‰‹–‹‘ˆ‘”‹‘‡–”‹…—ƒ—–Š‡–‹…ƒ–‹‘

ͶͺͶ

”ƒ˜‹ Ǥƒ”’ƒ–‡ǡƒ‰•‡Ǥƒ•‘†ǡƒ‡•ŠǤƒœƒ

 ͻͻǤ ƒ‡‡’ƒ”–—”‡ƒ”‹‰›•–‡•‹‰‘—‰Š”ƒ•ˆ‘”

ͶͻͲ

ƒ–‘•ŠǤƒ†—”‡ǡ”ƒ˜‹Ǥƒƒ™ƒ”

 

AuthorIndex               



















Ͷͻ͵

   

Big Data Analytics

Big Data Analytics

1

Cognitive Knowledge Engineering

2

Big Data Analytics

Chapter 1

Improvement in DRINA for Energy Efficient Wireless Sensor Networks Vaishali Sarode

K S Bahagat

Sachin Barhate

E&TC Dept., JT Mahajan Polytechnic, Faizpur, Maharashtra.

E&TC Dept., JT Mahajan COE, Faizpur, Maharashtra.

E&TC Dept., JT Mahajan Polytechnic , Faizpur, Maharashtra.

[email protected] [email protected] [email protected] ABSTRACT To perform error free monitoring dense wireless sensor networks are deployed in various categories of applications. Because of too huge network size, there are most probability of data redundancy i.e. multiple nodes may send similar information towards sink node. Data routing for in network aggregation (DRINA) techniques r e s o l v e s this problem o f data r e d u n d a n c y . This algorithm preserves energy by accomplishing data fusion and a g g r e g a t i o n . But as the nodes from overlapped paths it needs to be take care for the certain factors like utilization and allocation of memory, number of packets loss during transmission over the network etc. Thus energy exhaustion for those nodes gets influenced by these different f a c t o r s . Therefore proper balance must be maintained between maximization o f data aggregation and energy balance. Over data aggregation regardless of current state of node may lead to stoppage of some nodes leading to unstable network structure. Hence we archives data The aggregation by adaptive state aware algorithm. algorithm maximizes the possible data aggregation by proper formation and upgradation of Hop-Tree, takes the local state of nodes to build and maintain Hop-Tree, depends on Time-To-Live (TTL) mechanism to limit the Hop-Tree update range to avoid over-overlapping of paths according to the correlation of events, and balance the data load on the backbone nodes of Hop-Tree to further balance the energy consumption. Experimentation shows that our algorithm can maximize the possible data aggregation while balance the energy consumption among nodes and enhance the monitoring ability of WSNs significantly. Keywords Routing, Data Aggregation, Wi r el e ss sensor net- work, DRINA.

1.

INTRODUCTION

WIRELESS sensor network (WSN) is of a group of spatially disperse sensors nodes for monitoring the physical conditions of the environment like temperature, sound, pollution levels, pressure, etc. These networks have been used in different applications such as environmental or physical monitoring, land security, manufacturing systems, critical organization systems, etc. WSNs usually produce a large amount of information that needs to be routed across the networks in multi-hop fashion toward a sink or base station, which works as a gateway to monitoring center. Routing data is principal venture in the data gathering process. Since sensor nodes have very limited amount of energy generally battery

powered, the amount of gathered data significantly affects energy consumption. Thus, energy utilization, optimization and conservation is major key issue in WSN. Data fusion and Data aggregation techniques are used in order to save energy . A strategy to optimize the routing task for the available processing capacity can be provided by the intermediate sensor nodes along with the routing paths. Data aggregation is a technique of collecting the data from multiple sensor nodes to remove redundant transmission and provide fused information to the sink or base station. Due to the redundancy in raw data gathered by the sensor nodes, in-network aggregation can be used to decrease the communication cost by eliminating redundancy and forwarding only smaller aggregated data which intern decreasing the communication costs and energy consumption due to this the network lifetime is increases. Due to the dynamic nature of the WSN and the massive deployment of WSN to sense an event and report data and the characteristics of sensor nodes, coupled with the short range of communications, nodes must cooperate between themselves to finish a certain task simply if there is no cooperation between nodes there is no functioning network. For example if a sensor network is being deployed for target tracking purposes, nodes report to each other about sensed target and then to the other nodes in the routing path until reaches to the cluster head and the cluster head then reports to the base station. To get the sensed data to the cluster head from a sensor which is far away from the c luster head, cooperation must occur between nodes to be able to forward the data. And for cooperation to happen between nodes, a trust must exist, which means cooperation influences trust and vice versa. The expected contribution is building a probabilistic frame- work model to calculate and continuously update trust values between nodes in wireless sensor networks based on the sensed event and to exclude malicious and faulty nodes from the network. In other words, creating a framework to maintain the security and the reliability of a sensor network by examining the trust between nodes, so every node has a trust value for every other node in the surrounding area and based on that value the cooperation occurs between nodes. The principal challenges in routing algorithms for WSNs are to provide assured guarantee to the delivery of the sensed data for failures of nodes and overcome interruptions causing failure of network. As data packets contains aggregated data from various sensor nodes. Most of the time node failures become more critical in data aggregation performed along the routing paths. Whenever one of these packets is lost over the network, the amount of aggregated information will also be stray. In WSN, data aggregation techniques should provide certain features such as reduced number of messages for setting up a routing tree, maximum aggregation rate, maximizing number of overlapping routes, and reliable data transmission

3

Cognitive Knowledge Engineering

2. AN IMPROVEMENT ALGORITHM

IN

DRINA

The main goal of our Algorithm is to build an adaptive routing tree (Hop-Tree) for in-networking data aggregation according to local states and event correlation which can maximize the data aggregation while balance energy consumption, to enhance the monitoring capability of WSNs.

2.1 Hop Tree Configuration Phase In this work, a Hop-Tree rooted at Sink is formed after deploying sensor nodes, with the shortest paths (in hops) that connect all source nodes to the Sink while maximizing the possible data aggregation and balancing energy consumption. Algorithm 1 Hop Tree Configuration Phase start 1. Node sink sends broadcast of HCM message with the value of HTT=1 //Ru is the set of nodes that received the message HCM 2. foreach u ‫ ׫‬R(u) 3. do 4. if H T Tu > H T TH C M and F irst S endingu 6. N extH opu = IDH C M ; 7. H T Tu = H T TH C M + 1; 8. IDH C M = IDu ; 9. H T TH C M = H T Tu ); 10. StateH C M = Stateu ; 11. N odeu sends broadcast message of the HCM with the new values 12. F irstS endingu = false; 13. end 14. else if H T Tu == H T TH C M and StateN extH op(u) < StateH C M then 15. N extH opu = IDH C M ; 16. end 17. else 18. Node u discards received message HCM 19. end

2.2

Cluster Formation

When an event occurs, a cluster based on the nodes which detect it (we may call them event nodes) will be formed. The key process of the cluster formation phase shown in below Algorithm is the election of the leader node (called Coordinator) for the cluster. Input: S //set of nodes that detected event Ouput: u //A node of set S is elected as group leader Algorithm 2 Cluster Formation and Leader Selection 1. foreach u ‫׫‬ 2. do 3. role(u) = coordinator ; 4. // Node u sends message MCC in broadcast. Announcement of event detection; 5. // Nu is the set of neighbors of node u ‫ ׫‬S 6. foreach w ‫ ׫‬N(u) 7. do 8. if HTTu > HTTw then 9. roleu =collaborator; 10. Node u retransmites MCC message received from node w ; 11. end 12. else if HTT(u) == HTT(w) and ID(u) > ID(w) and State(HCM) > State(u) then 13. role(u) = collaborator; 14. Node u retransmites MCC message received from node w ;

15. 16. 17. 18. 19. 20.

end else Node u discards the MCC message received from w; end done done

2.3 Route Establishment and Hop Tree Update Each Coordinator, the elected group leader, starts establishing a path which will be the new part of the backbone of the HopTree. A Hop-Tree should be updated for gaining shortest paths connecting all source nodes while maximizing data aggregation and balancing energy distribution as events occur one by one. We update the HTT value of each node properly to achieve this goal. Algorithm 3 Route Establishment and Hop Tree Update 1. Leader node v of new event sends a message REM(Route establishment message) to its Next Hop(v) ; 2. repeat 3. //u is the node that received the REM message, that was sent by node 4. if u == Next Hop(v) then 5. if HTT(u) > 0 and Energy(Next Hop(v)) < Threshold Energy 6. then 7. HTT(u)= 0 8. //Node u is part of new route built 9. Role(u) = relay ; 10. Node u sends message REM to its Next Hop(u); 11. Node u broadcasts the message HCM with the value of HTT=1 12. Nodes that received HCM message sent by node u will run algorithm 1. 13. end 14. end Process continues until Find out the sink node or a node belonging to a routing structure already established;

2.4 Data Transmit-Retransmit Event coordinator v determine TTL of v to reach sink node. v determines d1 i.e. number of Hops to reach already established route (node u) u determines d2 i.e. number of Hops to reach sink node Algorithm 4 Data Transmit-Retransmit 1. If TTL > d1+d2 2. then 3. v diverts data to already established path (a node u), at every intermediate node route repair mechanism is applied 4. end 5. else 6. v follows shortest path to sink node, at every intermediate node route repair mechanism is applied 7. end

2.5 Route Repair Mechanism The Hop-Tree is unique, and any failure of nodes will cause disruption in data transmission. The route repair mechanism in the algorithm is a kind of piggybacked, ACK-based mechanism, and it may cause such problems as a longer delay due to several faulty nodes in the path and routing loops. A node sends data to its next hop and runs timer for arrival of acknowledgement from its next hop also a node buffers the data. If node receives acknowledgement from its next hop then it release buffer. Otherwise node determines best alternate next hop and retransmits the data from its buffer. 4

Big Data Analytics

3. SIMULATION RESULTS We compare our proposed algorithm i.e. improved DRINA and DRINA by taking consideration of 4 parameters ‡ Energy Depletion ‡ Event Loss Ratio

3.4 Number of Events Monitored by Sink Node Number of Events Monitored by Sink Node is the number of events that monitored by the Sink finally. Our proposed algorithm monitors more number of events as compared with DRINA

‡ Number of Lost Events ‡ Number of Events Monitored by Sink Node

3.1

Energy Depletion

Communication energy consumption in data gathering paths depends on the number of packets and a forwarding node receives and transmits. Thus energy depletion proved measure of energy consumed in network for forwarding data packets. From figure 1 it is clear that as the number of nodes increases the energy depletion also increases drastically. Our proposed algorithm consumes less energy with compared to DRINA algorithm. Thus providing better energy utilization for sensor nodes.

Figure.3. Comparision of DRINA and Prposed Algorithm in terms of Lost events

3.2 Event Loss Ratio It is the ratio of the number of events which are not monitored by the Sink to the total number of events. From figure 2 we can conclude that our proposed algorithm is more capable to monitor events with compared to DRINA. Our proposed algorithm monitors more events with compared to DRINA algorithm. Thus producing less Event loss ratio as compared to DRINA

3.3 Number of Lost Events Number of Lost Events is the number of events which are not monitored by the Sink. Form graph it is clear that our proposed algorithm can monitor more number of events thus number of lost events are small for proposed algorithm

Figure. 1. Comparison of DRINA and Proposed Algorithm in terms of Energy Depletion

Figure. 2. Comparison of DRINA and Proposed Algorithm in terms of Event Loss Ratio

Figure. 4. Comparison of DRINA and Proposed Algorithm in terms of Number of Events Monitored by Sink Node

4. CONCLUSION The need of deployment of wireless sensor network is increasing day by day. Aggregation helps in energy management and increasing efficiency of the network. Thus aggregation aware routing algorithms plays major role for WSNs. Each algorithm has its own set of advantages. Improvement of one parameter may results in lowering of other. For e.g. increasing aggregation will improve network lifetime but it may also increases delay in the network. Thus selection of routing algorithm and methodology is completely application specific. The main advantage of aggregation is improvement of network lifetime and efficiency. Depending on application a suitable aggregation technique should be selected which fulfils the requirement of the purpose of implementation of WSN. In this work we discussed various aspects of WSNs, routing, various routing challenges and design issues of wire- less sensor network. Also we overviewed DRINA Algorithm. Experimental analysis shows that DRINA algorithm provides better solution for aggregation while comparing other algorithms in various parameters like delay, latency. This paper addresses the problem of DRINA. Over- overlapping of paths leads to consumption of more memory thereby energy. Hence we propose improved DRINA algorithm. We propose tradeoff between data aggregation maximization and energy balance. Experimentation proves that proposed algorithm can maximize the possible data aggregation while balancing the energy consumption, making WSNs monitor more events. We proposed algorithm improved DRINA in order to have maximum aggregation and lower energy depletion. Taking consideration of parameters like Energy Depletion, Event Loss 5

Cognitive Knowledge Engineering Ratio, Number of Lost Events, Number of Events Monitored by Sink Node our algorithm out performs DRINA.

5. FUTURE SCOPE Many researches show that the energy spent on data transmission dominates the whole energy consumption for sensor nodes. Experimentation proves that proposed algorithm reduces energy consumption. In the upcoming Internet of Things (IoT) the everyday objects that surround us will become proactive actors of the Internet, generating and consuming information. It is predicted that the potential of the wireless sensor networks (WSN) paradigm will be fully unleashed once it is connected to the Internet, becoming part of the Internet of Things (IoT). Still in that case, again data will not be directly communicated to sink node. Thus it will again requires aggregation concept in order to utilize bandwidth. Thus aggregation algorithm in adaptive routing needs to be modified to stochastically select the nodes that will be the part of communication structure.

6. REFERENCES 1]Zhang Jiao, Ren Fengyuan, He Tao and Lin Chuang³'DWD Aggregation Protocol Based on Dynamic Routing in Wireless Sensor Networks,´ 2009 International Conference on

Communications and Mobile Computing. 2]Suat Ozdemir and Yang Xiao, ³6HFXre data aggregation in wireless sensor networks: A comprehensive overview,´ ELSEVIER Computer Networks 53 (2009) 20222037. 3]Giuseppe Anastasi, Marco Conti, Mario Di Francesco and Andrea Passarella³(QHrgy conservation in wireless sensor networks: A survey,´Ad Hoc Networks 7 (2009) 537568. 4]Bin Zhang, Wenzhong Guo, Guolong Chen and Jie Li ³,Q-network Data Aggregation Route Strategy Based on Energy Balance in WSNs,´2013 11th International Symposium and Workshops on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt), May 1317, 2013. 5]Jamal N. Al-Karaki, Raza Ul-Mustafa and Ahmed E. Kamal, ³'DWD aggregation and routing in Wireless Sensor Networks: Optimal and heuristic algorithms,´ Computer Networks 53 (2009) 945960. 6]Leandro Aparecido Villas, Azzedine Boukerche, Heitor Soares Ramos, Horacio A.B. Fernandes de Oliveira, Regina Borges de Araujo, and Antonio Alfredo Ferreira Loureiro, ³'5,1$ A Lightweight and Reli- able Routing Approach for InNetwork Aggregation in Wireless Sensor 1HWZRUNV´  IEEE TRANSACTIONS ON COMPUTERS, VOL. 62, NO.4, APRIL 2013.

6

Big Data Analytics

Chapter 2

Implementation of Division and Replication of Data in Cloud P.D. Patni

S.N.Kakarwal

P.E.S. College of Engineering, Aurangabad, Maharashtra, India

CSE Department, P.E.S. College of Engineering, Aurangabad, Maharashtra, India

[email protected]

[email protected] ABSTRACT In cloud system the data is outsourced on the cloud, this may create security issues. In this paper we propose Division and Replication of Data in Cloud (DRDC) which can take care of security issues without compromising the performance. In this system, file uploaded by the client is first encrypted then divided into fragments. Then these fragments are replicated over the cloud nodes. Fragmentation and replication is carried out in such a way that each node contains only a single fragment. Thus if any one of the node is intruded by hacker, no significant information is revealed, and thus security is maintained. To further increase the security, nodes are separated by T-coloring graph method. Due to the T-coloring, the effort needed by an attacker to breach the security is increased multiple times. In addition to this, in this paper we compare this system (DRDC) with other methodologies.

Thus both security and performance are of paramount importance for clouds. In division and replication of data in cloud (DRDC), both performance and security issues are mitigated. This system first encrypts the file then fragments the file into pieces and replicates them and placed at distinct nodes within the cloud. There is no meaningful information in individual fragment increasing the data security. Further, the attack on a single node does not reveal the locations of other fragments. In addition to this, to ensure further security, the selected nodes should not be adjacent. T-coloring is used for separation of nodes. Replication of fragments over the nodes that generate highest read/write request improves data retrieval time [3].

Keywords Cloud security, data fragmentation, replication, T-coloring.

1. INTRODUCTION The cloud computing concept has revolutionized the usage and management of the information technology world. This concept has many applications in various businesses and organizations. It is also useful for individual users. Despite its usefulness, it has many security concerns. Security concern is the main obstacle preventing the wide spread adoption of cloud computing. This security issues may be due to- The core tHFKQRORJ\¶VLPSOHPHQWDWLRQ90HVFDSHVHVVLRQ riding etc.)

Figure1: Division and replication of data in cloud (DRDC)

- Cloud Service offerings (structured query language injection, weak authentication schemes, etc.)

Figure 1 shows the working of division and replication of data in cloud (DRDC) and T-coloring. We are giving ten types of replication strategies as comparative techniques to this methodology (DRDC). These are-

- Cloud characteristics (data recovery issues, internet protocol issues, etc.) shared environment of cloud results in many security concerns [1]. The security of components of cloud ensures cloud security. To ensure highest level of security, the weakest component of system must be secured, because any weak entity can put the whole cloud at risk. So the security mechanisms should be such that it will be difficult for an attacker to retrieve data even after successful intrusion in the cloud. In addition, data leakage must also be minimized. For a large scale system, there is data replication strategy to deal with data retrieval time, data availability and reliability. But there are many numbers of nodes involved in replication strategy which increase the chances of attack [2].

A-star based replication strategies- i) DRPA-star, ii) weighted A star, iii) Aࣅ star, iv) suboptimal A star, v) suboptimal A star2, vi) suboptimal A star3. Then bin packing based techniques like vii) local min-min, viii) global min-min. Then greedy based techniques like ix) greedy algorithm and x) genetic replication algorithm. These techniques determine the number and locations of replicas and improve the performance. The data center network architecture mainly used in this study is three tiers. In this paper we proposed a DRDC system which provides the security to the cloud and also increases the performance. This scheme first encrypts the file. Then fragmentation and replication takes place. Nodal placement of fragments is done with the help of T-coloring. This scheme ensures that no meaningful 7

Cognitive Knowledge Engineering information revealed to the attacker even in case of successful intrusion of node. There is also control over replication of fragments, so that each of the fragments replicated only once so that performance is increased without decreasing security.

increase performance in this system, we have to compromise the security. To balance the security and performance DRDC methodology helps, as it does not store entire file on a single node. This system fragments the file and then replication of fragmented file is done. Thus even in a case of security breach, no significant information is revealed to an attacker. Replication is also in a controlled manner so that each fragmented file has only one replica, thus data security is not compromised in spite of maintaining performance [9].

2. LITERATURE SURVEY A technique presented by Juels et al. involves data migration to the cloud by iris file system. To maintain the integrity and freshness of data, a gateway application uses the merkle tree. In WKLV WHFKQLTXH WKH VHFXULW\ RI GDWD PDLQO\ GHSHQG XSRQ XVHU¶V employed scheme. This system is unable to prevent data loss or access by other VM [4]. M. Tu et al. presented a secure and optimal placement of data objects in a distribution system. The encryption key is divided and division is done through threshold secret sharing scheme. This scheme pays attention to the replication problem with security and access time improvement. In this scheme data files are not fragmented and are handled as a single file. This scheme mainly focuses on encryption key security unlike our methodology [5]. G. Kappes et al. LQ WKHLU ³'LNH 9LUWXDOL]DWLRQ-aware Access Control for Multitenant File V\VWHPV´ SUHVHQWDWLRQ DGGUHVVHV virtualized and multitenancy related problems. This architecture works by combining the native access control and the tenant name space isolation. But in this system there might be leakage of critical information due to inadequate sanitization or improper VM. DRDC methodology involves fragmentation of data file and multi nodal storage of single file and thus prevents leakage of critical information [6]. D. Zissis et al. LQ WKHLU SDSHU ³$GGUHVVLQJ FORXG FRPSXWLQJ VHFXULW\ LVVXHV´ DGYLVHV WKH XVH RI D WKLUG SDUW\ IRU VHFXULW\ RI cloud data. They use public key infrastructure, so that the level of trust is increased in the communication between the involved parties. Use of smart card was advised for the storage of the key at the user level. Tang et. al. also uses third party and public key cryptography for data security in cloud. But they do not use PKI infrastructure. The public key cryptography and the threshold secret sharing scheme are combined to protect the symmetric keys. However tampering and loss due to virtualization and multitenancy are not prevented [7]. 0D]KDU$OLHWDOLQWKHLUSDSHUµ'5236: Division and replication of data in FORXG IRU RSWLPDO SHUIRUPDQFH DQG VHFXULW\¶ DGYLVH fragmentation and replication of data and then use of T-coloring for multimodal placement of data. The nodes are selected on the basis of centrality measures that ensure an improved access time [8].

3. PROPOSED SYSTEM If in a cloud, file is stored on a single node, there is significant risk for data security. In this kind of system the retrieval time can be decreased by replicating the files at multiple nodes. But this increases the risk of data security multiple times. Thus in order to

7KHUH¶VDFORXGPDQDJHULQWKH'5'& system which is a secured entity. The cloud manager performs following functions ± -

Receiving the file Encryption of file by AES algorithm Fragmentation of file Nodal selection and each fragment assigned a single node with the help of T-coloring. Figure 2 shows the overall framework of DRDC method. Fragments replication and storage of each replicated fragments over separate cycle of nodes again with the help of T-coloring. The user can decide the fragmentation threshold in terms of size or percentage. The user can divide the file efficiently so that no significant information is there in a single fragment. If the user does not give details of fragmentation threshold then default percentage threshold is set and used while fragmenting the file. The cloud manager has to pay attention to the communication channel between cloud manager and client, and make sure that the channel is secure. As soon as file is divided into fragments, this system assigns the cloud nodes for each fragment. Centrality measures are employed to reduce the retrieval time. Three centrality methods are mainly used; these are betweenness, closeness and eccentric centrality. These measures may result in placement of fragments on adjacent nodes, thus compromises security. This is where the concept of Tcoloring is justified. In this concept a set T is built starting from zero to random positive number. To make this system work, colors are given to the nodeV /HW¶V FRQVLGHU WKHUH¶V DQ open_color before placing the fragment, as soon as fragment is placed on one of the node, then close_color is given to nodes surrounding the assigned node up to the T distance. This system makes cloud more secure, although somewhat performance is decreased due to less availability of central nodes. Further in order to increase performance or to decrease the access time, data replication is done in a controlled manner. Again while placing the replicated fragments; concept of T-coloring is used. In data replication, some of the fragments may not be replicated due to T-coloring due to less number of nodes. If client requests to download the uploaded file, then all the fragments are reassembled into a single file by cloud manager and then that file is sent to the client.

8

Big Data Analytics

Figure 2: DRDC Framework

3.1 AES Algorithm

i) Sub Bytes

Advanced Encryption Standard (AES) is one of the most secure encryption algorithms available.

ii) Shift Rows

7KHIHDWXUHVRI$(6DUHDVIROORZVí

iii) Add Round Key.

3.2 Fragmentation

-Symmetric key symmetric block cipher -128-bit data, 128/192/256-bit keys

Security of each node determines cloud security. An attack on one node makes easy for the intruder to attack on subsequent nodes in case of homogenous system.

-Stronger and faster than Triple-DES -Provide full specification and design details - Software implementable in C and Java ALGORITHM 1: Advance Encryption Standard [10]. 1. Key Expansions²round keys are derived from the cipher key using Rijndael's key schedule. AES requires a separate 128-bit round key block for each round plus one more.

But in heterogeneous system same effort will not result in intrusion of subsequent nodes. In this system attack on one node results in compromised security of information available only on that node, because in this system data file is fragmented and stored on different nodes. In addition to this, the possibility of finding fragments on all of the nodes is very less, if an attacker is not VXUHDERXWIUDJPHQW¶VORFDWLRQ. If number of nodes increases the probability of an intruder to obtain the data file decreases. If there are thousands of nodes in a cloud system, then that cloud system is relatively more secure.

2. Initial Round i) Add Round Key²each byte of the state is combined with a block of the round key using bitwise xor. 3. Rounds i) Sub Bytes²a non-linear substitution step where each byte is replaced with another according to a lookup table. ii) Shift Rows²a transposition step where the last three rows of the state are shifted cyclically a certain number of steps.

3.3 T-coloring In T-coloring graph vertices are colored. While coloring the vertices one thing is kept in mind that two adjacent vertices does QRW DSSHDU LQ RQH 7 ILHOG )RU H[DPSOH VXSSRVH µD¶ DQG µE¶ DUH two adjacent vertices, then coloring is done in sXFKDZD\WKDWµD¶ DQGµE¶GRHVQRWDSSHDULQ7ab, where Tab is a set of nonnegative integers associated to the edge [a, b]. This is called vertex coloring.

iii) Mix Columns²a mixing operation which operates on the columns of the state, combining the four bytes in each column. iv) Add Round Key 4. Final Round (no Mix Columns) 9

Cognitive Knowledge Engineering allocation, replication and further nodal allocation of replicas. When client requests a uploaded file for downloading then merging of fragments and decryption is done by cloud server. Data Centre Network (DCN) used for communicational purpose in cloud. There are many DCN architectures mainly three tier, Dcell, fat tree etc. We mainly use Three tier architecture in this study [11].

4. COMPARATIVE TECHNIQUES Figure 3. T-coloring node presentations. This methodology is compared with other methodologies like 6LPLODUO\WKHUH¶VDQHGJHFRORULQJLQZKLFKQRWZRDGMacent edges are of same color. i. ALGORITHM 2: Fragment placement by T coloring [8] Inputs and initialization: O = { O1, O««ON } O = {sizeof( O1 ), sizeof( O2 «VL]HRI2N ) } Col = {open_color,close_color} Cen = {cen1, cen««cenM,} Col ĸopen_color ‫ ׊‬i Cen ĸ ceni ‫ ׊‬i Compute: for each Ok ‫ א‬O do Select Si |Si ĸindexof ( max ( ceni ) ) If colSi =open_color and si >=ok then Si ĸ Ok si ĸ si - ok colSi ĸ close_color Si ¶ ĸ distance(Si,T) ColSL¶ ĸ close_color end if end for

ii.

iii.

3.4 Replication After fragmentation and nodal placement, fragments of file are replicated. Replication is in a controlled manner. So that only one replica of each fragment is made. Due to replication the access time is reduced hence performance is increased. Again nodal placement of replicated fragments is done with the help of T-iv. coloring. ALGORITHM3)RUIUDJPHQW¶VUHSOLFDWLRQ>@ For each Ok in O do select Si that has max (Rik + Wki) if colSi = open_color and si >=ok then Si ĸOk si ĸ si - ok colSi ĸ close_color Si¶ ĸmdistance(Si,T) ColSL¶ĸ close_color End if end for

v.

vi.

vii.

viii.

3.5 System Model i) Cloud client ± client may be data owner or data user.

ix.

Data owner is a person who uploads file on cloud. Data owner knows fragment placement with their node numbers. Data user is the person who downloads or view the files uploaded x. by others. Authentication is necessary for file accessing, otherwise he is considered as an intruder or attacker. ii) Cloud server ± When client uploads a file then cloud server performs many functions like encryption, fragmentation, nodal

DRPA-star ± This is A-star based technique, for the data replication problem. It starts from the root, called the start node. Intermediate nodes provide the partial solution and leaf nodes provide the complete solution. EaFK QRGH¶V FRPPXQLFDWLRQ FRVW LV JLYHQ E\ µFRVWQ  JQ KQ ¶ WKLV FRVW IRU D QRGH Q LV WKH estimated cost of the cheapest solution through n. Here g(n) is the search path cost from start node to the current node n and h(n), called the heuristic, is the lower bound estimate of the path cost from n to the solution. The DRPA-star searches all of the solutions of allocating a fragment to a node. The solution that minimizes the cost within the constraints is explored while others are discarded. The selected solution is inserted into a list called the OPEN list. This list is ordered in the ascending order so that the solution with the minimum cost is expanded first. WA-star ± This is a refinement of DRPA-star which uses weighted function to estimate the cost. It is given as f(n) = g(n) + h(n) + ࣅ >íGQ ' @KQ ZKHUHGQ LVWKHGHSWKRIQRGHQDQG D is the anticipated depth of the desired goal node. WA-star identifies solution within 1+ ࣅ of DRPA-star. Aࣅ-star- This is also an extension of DRPA-star. The searching is focused on a particular space such that search would not deviate from optimal solution by a factor ࣅ. This technique uses two lists OPEN and FOCAL. The FOCAL list is a sub-list of OPEN list and contains only those nodes that have f that do not deviate from lowest f by a factor greater than 1+ ࣅ. The node selection is done from the FOCAL list. Aࣅ star identifies a solution within a range of 1+ ࣅ of DRPA star. SA1 ± (suboptimal assignment) this is DRPA-star based heuristic. In SA1, at level R or below, only the best successors of node with the least expansion cost are selected. SA2- This is also a DRPA based heuristic. In SA2, when the depth level R is reached for the very first time, best successor with the minimum cost are selected. Other successors are discarded. SA3- Again a DRPA based heuristic. Discarding done similar to SA2 except that the nodes are removed from the OPEN list except the one with the minimum cost. LMM (Local min-min) ± Bin packing based technique. It sorts the file fragments based on the replication cost of the fragments. ,WDVVLJQVWKHIUDJPHQWVLQWKHDVFHQGLQJRUGHU,IWKHUH¶VDWLHWKH fragment with lower size is chosen. GMM (Global min-min) ± It chooses the fragment with global minimum of all the RC associated with the fragment. If minimum RC is same for two different fragments, then selection is done randomly. Greedy algorithm ± Firstly algorithm searches through all of the cloud nodes. Then node with the minimum replication cost is selected for a file fragment. Then in second iteration this algorithm finds second node with lower RC, which in conjunction with the site already picked in first iteration. GRA (Genetic replication algorithm) - This technique shows great stability in various scenarios. It mainly uses mix and match strategy. It consists of chromosomes representing various schemes for storing fragments over the nodes. In this technique selection, crossover and mutation operations are performed to select the best chromosome [12] 10

Big Data Analytics DRDC method has minimum response time and hence higher performance as compared to other technique as shown in figure 4.

5. RESULTS AND ANALYSIS If a single node stores all the information, then upon intrusion of WKDWQRGHWKHUH¶VDVHFXULW\ULVNRIDQHQWLUHGDWDILOH2QWKHRWKHU hand if only a fragment of file is store on a single node, then only that fragment is revealed upon intrusion. In our methodology, person has to intrude large number of nodes to obtain significant data. This is because in our methodology fragments are stored on distinct nodes with the help of T-coloring. For an attack to be successful the number of nodes which are intruded must be greater than n. 7KHUH¶VDQHTXDWLon which determine an effort done by a person to attack a node. ETotal = min(EAuth, n × ENode)

(1)

Where ETotal is effort necessary to breach the data; EAuth is an effort needed to break in authentication and ENode is effort needed to breach a single node. In DRDC methodology we mainly pay attention to the data security in cloud, so we neglect the effort needed to breach authentication. So here we can conclude that an effort needed by a person to breach the security is directly proportional to the number of fragments [13]. To measure the performance of the system, we mainly relied on Response time (RT) which is total network transfer time. It depends upon two factors mainly- time due to read request and time due to write request. It is calculated by RT =

Figure 4: RT versus No of Nodes

5.2 Effect of increasing number of fragment We studied the performance of these techniques by increasing number of fragments. Numbers of fragments selected were 10, 20, 40, 60, 80 and 100. Table 2 shows the RT values of the techniques with respect to increase in number of fragments. From this comparison we came to conclusion that at a given fragment number, the response time of our DRDC method is very less as compared to other heuristic techniques. Hence the performance of our DRDC methodology is more effective as shown in figure 5. Uploaded size of file taken for studying this effect is around 1Mb.

(2)

Where M is total number of nodes in the cloud, N is total number is the total read request and of file fragments to be placed, is total write request of kth fragment of file. We compared the results of DRDC system with the ten heuristic techniques. Study is done first by increasing number of nodes and secondly study is carried out by increasing number of fragments keeping the number of nodes constant. Values for ten heuristic techniques are directly taken from a paper mentioned in a reference. We used windows azure platform for our study. Comparison is done by comparing RT values of our system with the ten heuristic data replication strategies. File size taken for our study is between 10kb to 5Mb.

Figure 5: RT versus No. of file fragment

5.1 Effect of increasing number of nodes We studied the performance or RT by increasing the number of nodes. Numbers of nodes selected for simulation were 5, 25, 50, 100, 500 and 1024. RT values of our DRDC method and other ten heuristic techniques are given in table 1 against increasing number of nodes. Uploaded size of file taken for studying this effect is around 1Mb. From above results we conclude that our

11

Cognitive Knowledge Engineering Table1. RT versus No of Nodes No. of Nodes

Response Time (RT) %

5

DRPA 70

LMM 24

WA 71

GMM 37

$‫ڙ‬ 71

SA1 61

SA2 53

SA3 65

GREEDY 70

GRA 64

DRDC 24

25

75

30

70

42

70

57

49

64

69

65

13

50

73

35

71

47

71

60

42

62

70

65

13

100

74

40

72

50

72

60

42

62

70

65

17

500

74

40

72

50

72

60

42

62

70

65

19

1024

74

40

72 50 72 60 42 62 Table 2: RT versus No. of Fragments

70

65

21

10 20 40 60 80

DRPA 79 76 75 74 72

LMM 48 47 45 43 42

WA 76 74 70 68 64

GMM 55 55 52 50 48

100

72

41

62

46

No. of Fragments

Response Time (RT) % $‫ ڙ‬SA1 SA2 SA3 76 65 54 66 74 64 54 65 70 63 53 63 68 62 52 62 64 62 50 61 62

6. CONCLUSION In this paper we proposed a secured system for storage of data in cloud that is also very good in performance. The file which is uploaded on the server first encrypted, and then fragmentation and replication of fragments takes place. Nodes are assigned to the fragments and replicas with the help of T-coloring. Fragments are placed over the nodes in such a way that no node contains more than one fragment. This system of fragmentation and Tcoloring increases the effort of an attacker to intrude the system. Even in case of successful attack on a fragment, no significant information is revealed to an attacker. Lastly we compared the performance of our method with the ten heuristic replication strategies.

7. REFERENCES [1] K. Hashizume, D. G. Rosado, E. Fernndez-Medina, and E. B. Fernandez, 2013, ³$QDQDO\VLVRIVHFXULW\LVVXHVIRUFORXG FRPSXWLQJ´ -RXUQDO RI ,QWHUQHW 6HUYLFHV DQG $pplications, Vol. 4, No. 1, pp. 1-13. [2] L.M. Kaufman, 2009, ³'DWD VHFXULW\ LQ WKH ZRUOG RI FORXGFRPSXWLQJ´,(((6HFXULW\Dnd Privacy, Vol. 7, No. 4, pp. 61-64. [3] W. K. Hale, 1980, ³)UHTXHQF\DVVLJQPHQW7KHRU\DQG DSSOLFDWLRQV´ 3URFHHGLQJV RI Whe IEEE, Vol. 68, No. 12, pp. 1497-1514. [4] A. Juels and A. Opera, 2013, ³1HZ DSSURDFKHV WR security anGDYDLODELOLW\ IRUFORXGGDWD´ Communications of the ACM, Vol. 56, No. 2, pp. 64-73. [5] D. Zissis and D. Lekkas, 2012 ³$GGUHVVLQJ FORXG FRPSXWLQJ VHFXULW\ LVVXHV´ )XWXUH *HQHUDWLRQ &RPSXWHr Systems, Vol. 28, No. 3, pp. 583-592. [6] G. Kappes, A. Hatzieleftheriou, and S. V. Anastasiadis, 2013, ³'LNH9LUWXDOL]DWLRQ-aware Access Control for Multitenant

60

48

60

GREEDY 77 75 73 72 70

GRA 68 67 66 65 63

DRDC 25 24 33 35 35

68

62

36

)LOHV\VWHPV´ 8QLYHUVLW\ RI ,RDQQLQD *UHHFH 7HFKnical Report No. DCS2013-1. [7] Y. Tang, P. P. Lee, J. C. S. Lui, and R. Perlman, 2012, ³6HFXUH RYHUOD\ FORXG VWRUDJH ZLWK DFFHVV FRQWURO DQG DVVXUHG GHOHWLRQ´ ,((( 7UDQVDFWLRQV RQ 'HSHQGDEOH DQG 6HFXUH Computing, Vol. 9, No. 6, (Nov. 2012), pp. 903-916. [8] K. Bilal, S. U. Khan, L. Zhang, H. Li, K. Hayat, S. A. Madani, N. Min-Allah, L. Wang, D. Chen, M. Iqbal, C. Z. Xu, and A. Y. Zomaya, 2015, ³DROPS: Division and Replication of Data in Cloud for Optimal Performance and Security´ Concurrency and Computation: Practice and Experience, Vol. 25, No. 12, pp. 1771-1783. [9] A. Mei, L. V. Mancini, and S. Jajodia, 2003 ³6HFXUH dynamic fragment and replica allocation in large-scale distributed ¿OH V\VWHPV´ ,((( 7UDQVDFWLRQV RQ 3DUDOOHO DQG 'LVWULEXWed Systems, Vol. 14, No. 9, pp. 885-896. [10] Joan Daemen, Vincent RijPHQ  ³The Design of Rijndael: AES ± The Advanced EncU\SWLRQ6WDQGDUG´6SULQJHU ISBN 3-540-42580-2. [11] K. Bilal, S. U. Khan, L. Zhang, H. Li, K. Hayat, S. A. Madani, N. Min-Allah, L. Wang, D. Chen, M. Iqbal, C. Z. Xu, and A. Y. Zomaya, 2013, ³4XDQWLWDWLYHFRPSDULVRQVRIWKHVWDWH of the art data center architHFWXUHV´ &RQFXUUHQF\ DQG Computation: Practice and Experience, Vol. 25, No. 12, pp. 17711783. [12] S. U. Khan, and I. Ahmad, 2008, ³&RPSDULVRQ DQG analysis of ten static heuristics-based Internet data replication WHFKQLTXHV´-RXUQDORI3DUDOOHODQG'LVWULEXWed Computing, Vol. 68, No. 2, pp. 113-136. [13] J. J. Wylie, M. Bakkaloglu, V. Pandurangan, M. W. Bigrigg, S. Oguz, K. Tew, C. Williams, G. R. Ganger, and P. K. Khosla, 2001 ³6HOHFWLQJ WKH ULJKW GDWD GLVWULEXWLRQ VFKHPH IRU D VXUYLYDEOH VWRUDJH V\VWHP´ &DUQHJLe Mellon University, Technical Report CMU-CS-01-120,(May 2001).

12

Big Data Analytics

Chapter 3

Fast Apriori Algorithm for Frequent Itemset Mining Ratnadeep R. Deshmukh

Shubhangi D. Patil

Department of Computer Science and IT,

Department of Information Technology,

Dr. B A M U, Aurangabaad.

rrdeshmukh.csit@ba mu.ac.in

Government Polytechnic, Jalgaon

Luigi Troiano

Soumya Banerjee

University of Sannio, Benvento, Italy troiano@unisannio .it

Birla Institute of Technology Extension Center, Deoghar

soumyabanerjee @bitmesra.ac.in

Ajay R. Deshmukh Dept. of IT, PICT, Pune.

ajaydeshmukh14 [email protected]

sdkirange@gm ail.com ABSTRACT² Determining frequent objects is one of the most important fields in data mining. The adaptive Apriori algorithm proposed can be efficient in terms of time required than the Apriori algorithm. It is well known that the size of the database for defining candidates has great effect on running time and memory need. We presented experimental results, showing that the proposed algorithm always outperform Apriori. To evaluate the performance of the proposed algorithm, ZHKDYHWHVWHGLWRQ7XUNH\VWXGHQW¶VGDWDEDVHDVZHOODVDUHDO time dataset.

1. INTRODUCTION In Data Mining, Association Rule Mining is a standard and well researched technique for identifying the most promising relations between variables in large databases. Association rule is used as a precursor to different Data Mining techniques like classification, clustering and prediction. Association rule mining, which was introduced by Agrawal et al. [1], has become a popular research area due to its applicability in various fields such as market analysis, forecasting and fraud detection. Given a market basket dataset, association rule mining discovers all DVVRFLDWLRQ UXOHV VXFK DV ³$ FXVWRPHU ZKR EX\V LWHP X, also buys item Y DWWKHVDPHWLPH´7KHVHUXOHVDUHGLVSOD\HGLQWKH form of X->Y where X and Y are sets of items that belong to a transactional database. Support of association rule X -> Y is the percentage of transactions in the database that contain X U Y . Association rule mining aims to discover interesting relationships and patterns among items in a database. It has two steps; finding all frequent itemsets and generating association rules from the itemsets discovered. Itemset denotes a set of items and frequent itemset refers to an itemset whose support value is more than the threshold described as the minimum support. Since the second step of the association rule mining is straightforward, the general performance of an algorithm for mining association rules is determined by the first step [2]. Therefore, association rule mining algorithms commonly concentrate on finding frequent itemsets. For this reason, in this SDSHU ³DVVRFLDWLRQ UXOH PLQLQJ DOJRULWKP´ DQG ³IUHTXHQW LWHPVHWPLQLQJDOJRULWKP´WHUPVDUHXsed interchangeably. Apriori and FP-Growth are known to be the two important algorithms each having different approaches in finding frequent itemsets [3,4]. The Apriori Algorithm uses Apriori Property in order to improve the efficiency of the level-wise generation of frequent itemsets. On the other hand, the drawbacks of the algorithm are more time required for candidate generation and multiple database scans. FP-Growth comes with an approach

that creates signatures of transactions on a tree structure to eliminate the need of database scans and outperforms compared to Apriori. A recent algorithm called Matrix Apriori which combines the advantages of Apriori and FPGrowth was proposed [5]. The algorithm eliminates the need of multiple database scans by creating signatures of itemsets in the form of a matrix. The algorithm provides a better overall performance than FP-Growth [6]. Although all of these algorithms handle the problem of association rule mining, they ignore the time required for frequent itemset mining [7]. While performing the association rule mining using Apriori algorithm, the first level candidate itemset are generated and then these are used to generate the second level candidate itemset and so on. The number of database scans as well as time required for frequent itemset mining is more in this case. The solution to this problem is adaptive itemset mining in which the idea is to generate the frequent itemset for the last level, then find all transactions contained in the last level. These transactions are copied for all lower levels also and new transaction database is created by deleting the already copied transactions for each subsequent level. In this paper, a new approach for frequent itemset mining based on Adaptive Apriori Algorithm is proposed and compared with Apriori algorithm. In this paper a novel approach is proposed to improve the Apriori algorithm through the creation of intermediate dynamic database using MATLAB, where the database transactions are saved for each level separately. Thus repeated scanning is avoided and particular rows & columns are extracted and perform a function on that rather than scanning entire database. Results can be easily visualized and interpreted using graphical form display, The proposed adaptive approach showed a very good result in comparison to the traditional Apriori algorithm because there is a pruning process to those transactions in the database whose item count is less than minimum support. Hence the size of the database at each level reduces drastically which saves a lot of time, and a noticeable improvement in the speed by reducing the redundant scanning of the database.

2. RELATED WORK Apriori algorithm is a classical algorithm in the association rule mining field, the main idea is to use a low-dimensional frequent itemsets to obtain high-dimensional frequent itemsets through iterative step by step. However, the algorithm needs to scan the database frequently, which results in the I/O overburdened; the process that producing k-frequent itemsets by the (k-1)-frequent itemsets is too large, which leads to low efficiency; the 13

Cognitive Knowledge Engineering algorithm does not remove the transaction when the database FRQWDLQV WUDQVDFWLRQV WKDW QHHGQ¶W VFDQ 0DQ\ VFKRODUV KDYH been improved and optimized the Apriori algorithm from different angles by now.

step helps in filtering out candidate item-sets whose subsets (prior level) are not frequent. This is based on the antimonotonic property as a result of which every subset of a frequent item set is also frequent.T hus a candidate item set which is composed of one or more infrequent item sets of a prior level is filtered(pruned) from the process of frequent itemset and association mining.[4]

The DHP algorithm proposed by Park et al. [8] adopted the idea of dynamic hash hashing algorithms and pruning algorithm, and H[FOXGHWUDQVDFWLRQVWKDWFDQ¶WJHQHUDWHIUHTXHQWLWHPVHWVZKHQ traversing the database, then improved the efficiency of mining frequent itemsets.

Apriori Algorithm Input D, a database of transactions

Brin proposed a dynamic itemsets counting algorithm DIC [9], which can effectively reduce the frequency of scanning the database. This algorithm divides the transaction database into the same size, and accesses the data blocks in turn to generate the 1-frequent itemsets then generate the candidate 2-frequent itemsets by self-join, at last it merges the 1-frequent itemsets and candidate 2-frequent itemsets that each

Min_sup, the minimum threshold support

Ck Set of Candidate k-itemsets. Method: 1. L1 =Frequent items of length 1.

block generated, then repeat it until there is no new itemsets or reach the limit.

)RUN /N ɮN GR 3. Ck+1=candidates generated from Lk.

The algorithms above can work well when the amount of data is small and the dimension is not high, but in the environment of ELJ GDWD RU WKH GLPHQVLRQ LV KLJK WKHVH DOJRULWKPV FDQ¶W ZRUN well. Google came up with the MapReduce [10] framework in 2004, then Hadoop based on MapReduce becomes popular, and cluster-based parallel data mining gains more attention, research and application. A lot of classical data mining algorithms are parallelized on Hadoop. Based on the Apriori algorithm, Lin et al proposed several parallel algorithms SPC, FPC and DPC [11] based on MapReduce, the SPC algorithm assign the data set to all Map nodes, and execute mining operation in parallel, then in the reduce phase execute combining operation, the algorithm starts the Map and Reduce tasks only once, however the FPC and DPC algorithms require start MapReduce tasks repeatedly, determined by the dimension of frequent itemsets mining; Li et al proposed a parallel frequent itemsets mining algorithm PApriori [12], in the Map stage scan the transaction database to count candidate frequent itemsets, and perform statistical operations to get frequent itemsets in the reduce phase, but it also need to repeatedly start MapReduce tasks. S.Hammoud [13] proposed a parallel in each round of the iterative process, assigning slice datasets to each Map node and statistical candidate frequent itemsets, then merging to get frequent itemsets in the Reduce phase. Because of the MapReduce IUDPHZRUN¶V KLJK ODWHQF\ DQG ODFN RI LWHUDWion, Apriori DOJRULWKPGRHVQ¶WILWWKH0DS5HGXFHIUDPHZRUNZHOO6SDUNLVD memory-based parallel computing framework, and it can greatly improve the real-WLPH GDWD SURFHVVLQJ DQG HQVXUH WKH FOXVWHU¶V high fault tolerance and high scalability [14] in big data environments. Qiu H et al introduces a parallel Apriori algorithm based on Spark YAFIM [15], and the paper realized the classical Apriori algorithm in Spark environment, and the results showed that the algorithm based on Spark has greatly improved in performance than the algorithm based on Hadoop.

3. THE APRIORI ALGORITHM 3.1

Apriori Algorithm

One of the first algorithms to evolve for frequent itemset and Association rule mining was Apriori. Two major steps of the Apriori algorithm are the join and prune steps. The join step is used to construct new candidate sets. A candidate itemset is basically an itemset that could either be. frequent or infrequent with respect to the support threshold. Higher level candidate itemsets (Ci) are generated by joining previous level frequent itemsets are Li-1 with itself. The prune

Lk Maximal frequent itemsets in D

Output

4. For each transaction t in database D do. 5. Increment the count of all candidates in Ck+1 that are contained in t. 6. Lk+1 =candidates in Ck+1 with minimum support 7. end do 8. Return the set Lk as the set of all possible frequent itemsets

The main notation for association rule mining that is used in Apriori algorithm is the following. 1) A k ±itemset is a set of k items. 2) The set Ck is a set of candidte k-itemsets that are potentially frequent. 3) The set Lk is a subset of Ck and is the set of k-itemsets that are frequent. Drawbacks of Apriori Algorithm 1.

The Apriori algorithmic program takes longer time for candidate generation technique.

2.

The Apriori algorithmic program needs many scans of the database.

3.

Many trivial rules are derived and it will be hard to extract the most interesting rules.

4.

Rules can be inexplicable and fine grained.

5.

Redundant rules are generated.

3.2

Adaptive Apriori Algorithm

Our Adaptive Apriori implementation, proposes overcoming some of the weaknesses of the Apriori algorithm by reducing the size of the database at each level. This algorithm uses a dynamic technique to reduce the time required for candidate itemset generation. It is claimed that the size of the database is reduced at each level starting from last to first and hence the time required for candidate itemset generation is reduced as compared with basic Apriori algorithm. Here we generate the dynamic intermediate database for each level separately. For example, when scanning each transaction in the database find all the transactions which contains all items. These 14

Big Data Analytics transactions are considered for generation for level K itemset. The same transactions are also considered for generating level 1 to level k-1 itemsets. So these are copied for all these levels. Now the database is updated by deleting all these transactions. This updated database is again considered for generating level L k-1 itemsets. Hence the size of the database is reduced at level of candidate itemset generation as well as the time required is also minimized. Algorithm:

itemset. Again modify the intermediate database from level 1:K1and remove the related transactions from the database. At each step of candidate itemset generation we get the reduced database which helps to improve the running time of the algorithm. TABLE I THE TRANSACTIONS

Sr No.

Transact ion ID

Itemsets

Input D, a database of transactions Min_sup, the minimum threshold support Lk Maximal frequent itemsets in D

Output

Ck Set of Candidate k-itemsets. Method: 1.

Generate the Intermediate Database

a.

Find all transactions to be considered for level K containing all itemsets.

b.

Copy these transactions in the database to be considered for level 1 to K

c.

Delete these transactions from the database D and update the database.

d.

Now consider this updated (reduced in size) database for finding all transactions for level 1 to K-1

e.

Repeat the steps subsequently and update the database.

2.

Consider this updated database for candidate itemset generation at each step.

3.

L1 =Frequent items of length 1.

4.

)RUN /N ɮN GR

5.

Consider D as intermediate updated database for level k

6.

Ck+1=candidates generated from Lk.

7.

For each transaction t in database D do.

8.

Increment the count of all candidates in Ck+1 that are contained in t.

9.

Lk+1 =candidates in Ck+1 with minimum support

1

T1

Milk, Cheese

2

T2

Milk, Coffee, Butter

3

T3

Jam, Bread

4

T4

Bread, Butter, Cheese

5

T5

Coffee, Milk

6

T6

Milk, Bread, Butter, Jam

7

T7

Milk, Bread, Butter, Jam, Cheese,Coffee

Now the transaction database is scanned to find the transactions that can be used for generating frequent itemsets at level K=6. Such transaction is T7 in the above example which contains all the items. So this transaction is removed from the dataset and copied for intermediate database level 1:K.-1 (1:5) Similarly intermediate database is constructed for all levels as shown in table II to VIII TABLE II INTERMEDIATE DATABASE FOR LEVEL K=6

Sr No.

7

Transact ion ID

T7

10. end do

3.3 An Example of Adaptive Apriori Algorithm Assume that a large supermarket tracks sales data by VWRFNNHHSLQJ XQLW 6.8  IRU HDFK LWHP VXFK DV ³EXWWHU´ ³EUHDG´ ´MDP´ ´FRIIHH´ ´FKHHVH´ ´PLON´ LV LGHQWLILHG E\ D numerical SKU. The supermarket has a database of transactions where each transaction is a set of SKUs that were bought together. Let the database of transactions consist of following itemsets: The transaction set as shown in Table 1. Firstly, scan all transactions which are required to get frequent K itemset. Copy all these transactions from database to the intermediate database for level 1 to K. Remove all these transactions from the original database. Now again scan the new updated database to find the transactions that are needed to build frequent K-1

Milk, Bread, Butter, Jam, Cheese, Coffee

TABLE III

11. Return the set Lk as the set of all possible frequent itemsets In this algorithm, the intermediate database is generated to reduce the time required for candidate itemset generation. When the support count is established the algorithm determines the frequent itemsets. It generates the candidate itemsets as like the Apriori algorithm.

Itemsets

INTERMEDIATE DATABASE FOR LEVEL K=5

Sr No.

7

Transact ion ID

T7

Itemsets

Milk, Bread, Butter, Jam, Cheese, Coffee

TABLE IV INTERMEDIATE DATABASE FOR LEVEL K=4

Sr No.

6

Transact ion ID

T6

Itemsets

Milk, Bread, Butter, Jam 15

Cognitive Knowledge Engineering 7

T7

Milk, Bread, Butter, Jam, Cheese,Coffee

TABLE V INTERMEDIATE DATABASE FOR LEVEL K=3

Sr No.

Transact ion ID

Itemsets

count and the transactions ids that contain these items, and and then eliminate the candidates that are infrequent or their support are less than the min_ sup as shown in table 2. The frequent 1itemset is shown in table 3 The sets which are in bold will be deleted in frequent 2_itemset as shown in table 4. The sets which are in bold will be deleted in frequent 3_itemsets as shown in table 5. Table VIII shows the candidate itemset generated at level 1 using intermediate database for level 1 TABLE VIII

2

T2

Milk, Coffee, Butter

4

T4

Bread, Butter, Cheese

6

T6

Milk, Bread, Butter, Jam

7

T7

Milk, Bread, Butter, Jam, Cheese,Coffee

CANDIDATE I ITEMSET

Sr No.

Items

Support

1

Milk

5

2

Cheese

3

3

Coffee

3

4

Bread

4

5

Butter

4

6

Jam

3

TABLE VI INTERMEDIATE DATABASE FOR LEVEL K=2

Sr No.

Transact ion ID

Itemsets

1

T1

Milk, Cheese

2

T2

Milk, Coffee, Butter

3

T3

Jam, Bread

4

T4

Bread, Butter, Cheese

5

T5

Coffee, Milk

6

T6

Milk, Bread, Butter, Jam

7

T7

Milk, Bread, Butter, Jam, Cheese,Coffee

Table IX shows frequent 2 itemset generated using the intermediate table for level K=2

TABLE VIII CANDIDATE I ITEMSET

Sr No.

Sr No.

Items

Support

1

Milk,Cheese

2

TABLE VII

2

Milk, Bread

2

INTERMEDIATE DATABASE FOR LEVEL K=1

3

Milk, Butter

3

4

Milk, Jam

2

5

Cheese,Bread

2

6

Cheese,Butter

2

7

Cheese, Jam

1

8

Bread, Butter

3

9

Bread, Jam

3

10

Butter, Jam

2

Transact ion ID

Itemsets

1

T1

Milk, Cheese

2

T2

Milk, Coffee, Butter

3

T3

Jam, Bread

4

T4

Bread, Butter, Cheese

5

T5

Coffee, Milk

6

T6

Milk, Bread, Butter, Jam

7

T7

Milk, Bread, Butter, Jam, Cheese, Coffee

Now to generate the frequent 3 itemset, we need to consider only the intermediate database created for level 3. As can be noted here the size of the database is reduced to 4 only. Thus the proposed adaptive Apriori algorithm helps to improve the running time of the program.

Now by considering this newly constructed intermediate database at each level, firstly, scan all transactions to get frequent 1-itemset l1 which contains the items and their support 16

Big Data Analytics

4. ANALYSIS AND EVALUATION OF THE ADAPTIVE APRIORI Apriori Algorithm used to scan the database thrice but this paper presents an improvement on it by using adaptive algorithm and the concept of intermediate database generation. The analysis shows that the time consumed in improved Apriori in each group of transaction is less than the original Apriori. The memory space is reduced by using the dynamic database approach which partitions the original database initially and select one particular database out of this. It is an improvement as earlier the algorithm took exponential time but now it is reduced greatly.

4.1 Analysis and Evaluation of the Adaptive Apriori for execution time The second experiment compares the execution time original Apriori, and our improved algorithm by applying the Turkiye VWXGHQW¶V HYDOXDWLRQ GDWDEDVH IRU IDFXOW\  LQ WKH implementation. The 100 transactions considered here for analysis. The result is shown in Figure 1. Figure 1 shows that the improved Apriori reduce the execution time from the

Fig. 2. Execution time for different values of support for 7XUNL\HVWXGHQW¶VHYDOXDWLRQGDWDEDse for faculty 2

5. ANALYSIS AND EVALUATION OF THE ADAPTIVE APRIORI FOR REAL TIME DATA We have gathered the real time faculty evaluation data from 393 students of faculties from J T Mahajan College of Engineering, Faizpur. We have compared the execution time of original Apriori, and our improved algorithm by applying the real time database implementation. Table IX shows that the improved adaptive apriori reduces the execution time as compared to original apriori for real time dataset. TABLE IX EXEXUTION TIME TIME FOR REAL TIME DATASET

Sr No.

Support

Execution Time Apriori

1 2 3 4 5 6 7 8 9 10

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

2764.4949 2776.5642 2789.4587 2773.9934 2768.8197 2771.1176 2819.0367 2044.3822 56.3806 0.24622

Adaptive Apriori 2166.3458 2164.0938 2162.6268 2166.7127 2167.2221 2187.1428 2182.2447 1750.3789 58.8522 0.99224

6. CONCLUSION Fig. 1. Execution time for different values of support for 7XUNL\HVWXGHQW¶VHYDOXDWLRQGDWDEDVHIRUIDFXOW\

original Apriori. Figure 2 shows that the improved Apriori reduce the execution time from the original Apriori for Turkey faculty 2 dataset.

Association rule mining aims to discover interesting patterns in a database. There are two steps in this data mining technique. The first step is finding all frequent itemsets and the second step is generating association rules from these frequent itemsets. Association rule mining algorithms generally focus on the first step since the second one is direct to implement. Although there are a variety of frequent itemset mining algorithms, they are time consuming. The solution to this problem is adaptive frequent itemset mining. In this paper, adaptive frequent itemset mining algorithm, called Adaptive Apriori is proposed. The main advantage of the algorithm is avoiding the entire database scan at each level of candidate itemset generation. The proposed adaptive Apriori DOJRULWKP LV HYDOXDWHG RQ WKH 7XUNL\H VWXGHQW¶V HYDOXDWLRQ dataset and found to be most time efficient as compared to basic Apriori algorithm.

7. REFERENCES [1] $JUDZDO 5 7 ,PLHOLQVNL DQG $ 6ZDPL   ³0LQLQJ DVVRFLDWLRQ UXOHV EHWZHHQ VHWV RI LWHPV LQ ODUJH GDWDEDVHV´ ,Q Proceedings of the 1993 ACM SIGMOD International Conference on Management of Data, SIG02'¶1HZ /etc/dhcp/dhcpd.conf

prompt 1

7. Make sure the file looks as follows.

kernel /vmlinuz-2.6.32-431.el6.x86_64

[root@example ~]# cat /etc/dhcp/dhcpd.conf

initrd /disklessinitrd.img rw root=dhcp root=nfs:192.168.0.254:/diskless selinux=0 enforcing=0

default-lease-time 600;

menu

under

[root@example ~]# cat /var/lib/tftpboot/pxelinux.cfg/default

lable project

subnet 192.168.0.0 netmask 255.255.255.0 {

13. The installed system doesn't have any user, even root user doesn't have any password, so replace /diskless/etc/passwd and /diskless/etc/shadow with /etc/passwd and /etc/shadow

range 192.168.0.5 192.168.0.15;

[root@example ~]# cp -f /etc/shadow /diskless/etc/shadow

option domain-name "example.org";

[root@example ~]# cp -f /etc/passwd /diskless/etc/passwd

option routers 192.168.0.254;

14. Disable the firewall

next-server 192.168.0.254;

[root@example ~]# iptables -F

filename "pxelinux.0";

15. Now, boot the client with PXE boot. It will boot.

}

4.1 COMPARATIVE ANALYSIS

max-lease-time 7200;

8. Restart the DHCP service and chkconfig that on. [root@example ~]# service dhcpd restart;chkconfig dhcpd on

Table: 1 Comparative Analysis Characteristics

Thin Client

Thick Client

Diskless Client using open source Architecture

And here its mentioned disabled = yes , make it disabled = no , save changes and exit and restart the service

Low Watt Power Supply

Good

Poor

Good

[root@example ~]# service xinetd restart;chkcofig xinetd on

Energy Efficiency

Good

Poor

Good

Check the serverargs from /etc/xinetd.d/tftp file, by default, its /var/lib/tftpboot/

CPU Capacity

Poor

Good

Good

10. Configure NFS server

Cost

Medium

Poor

Excellent

[root@example ~]# vi /etc/exports

Availability

Medium

Good

Good

/diskless *(rw,sync,no_root_squash)

Support

Good

Good

Good

9. Configure tftp service. [root@example ~]# vi /etc/xinetd.d/tftp

34

Big Data Analytics

We have checked the performance of diskless client using open source architecture along with existing technologies available in the market, we can connect more than 1000 systems where as other technologies having some limitation while connecting more than 100 systems, we have calculated the performance of the technologies such as thin client, thick client and diskless client supported with open source architecture as per the characteristics shown in the table 1.1 Many of the aspect related with the speed, Availability, Graphics, Energy our solution is good. The most important part to understand in the POC is the dracut network image. If the root partition is on a network drive, one has to have the network dracut modules installed to create a network aware init image. This initramfs.img gets downloaded using tftp protocol and then creates an environment in Physical Memory in order to create a feasible environment for mounting the further file systems. Why we are promoting open source because we all know paid operating systems cost is not affordable to any section like school, colleges, universities, offices, industries etc. and this cost is increasing day by day, we need have a stable operating system and stable hardware cost also, but the scenario is different the cost of hardware is also not affordable, if we installed this solution to above said area it will save cost energy and affordable to the mass education and every on can dream for free operating systems and every one can happy to learn the computer system without any trouble.

5. CONCLUSION This research designed a Open source diskless technology using virtualization technique with switchover operating system architecture and it is one of the unique way to implement the diskless technology for mass education such as schools, Govt. offices, colleges, industries for cost effectiveness. More than 1000 of system can connect with single network and access the entire open source operating system at client end. Minimizes the services of hardware network engineers, how need to maintenance of single server machine. Resource can be

managed from central location where all client systems are connected through cloud. Three operating systems can be accessed using virtual environment with the help of cloud data center all over the world. As the diskless client images can be transferred over the network, this will help out for Decentralized System Management as well as Centralized System Control.

6. REFERENCES [1] Marisol Garcia-Valls, Tommaso Cucinotta, Chenyang Lu, 2014. Challenges in real-time virtualization and predictable cloud computing, Journal of System Architecture 60 726-740. @2014 Elsevier B. V. All rights reserved. [2] Kulthida phapikhor, suchart khummanee, panida songram, chatklaw jareanpon, 2012. Performance Comparison of the Diskless Technology, 10 th Internaional Joint Conference on Computer Science and Software Engineering (JCSSE). [3] Shingo Takada, Akira Sato, Yasushi Shinjo, Hisashi Nakai, Akiyoshi Sugiki and kozo Itano,´#,((($33 Approach to Scalable Network-Booting, Third International Conference on Networking and Computing. [4] 1RNL7DQLGD.HL+LUDNL0DU\,QDED³(IILFLHQWGLVNto- disk copy through long-distance high-speed networks with EDFNJURXQG WUDIILF´, Fusion Engineering and Design, www. elsevier.com/locate/fusengdes. [5] @

Hotels and Restaurants Reviews

Accuracy- 79.4%

57

Cognitive Knowledge Engineering Joint Sentiment-Topic (JST) Model [11]

Movie Reviews

AccuracyWithout prior information- 59.8 % Paradigm words- 74.2 % Paradigm words + -79.5%

Aspect-Sentiment Unification Model (ASUM) [10]

Electronics Product Reviews and Restaurants Reviews

Accuracy- 83 % to 85 %

Factorized LDA (FLDA) [15]

Product Reviews

AccuracyFor cold- 70 % to 74 % For Non-Cold- 83% to

Factorized Latent Aspect Model (FLAME) [18]

Hotels Reviews

91 %

AccuracyRMSE-0.98 Pearson Correlation inside reviews (ȡA)-0.195 Pearson Correlation between personalized ranking of Items ȡI - 0.333 Zero-One Ranking loss L0/1-0.196

6. REFERENCES [1] Liu, Bing. "Sentiment analysis and opinion mining." Synthesis Lectures on Human Language Technologies 5, no. 1 (2012): 1167. [2] Abbasi Moghaddam, Samaneh. "Aspect-based opinion mining in online reviews." PhD diss., Applied Sciences: School of Computing Science, 2013. [3] K. Schouten and F. Frasincar, "Survey on Aspect-Level Sentiment Analysis," in IEEE Transactions on Knowledge and Data Engineering, vol. 28, no. 3, pp. 813-830, March 1 2016. [4] http://en.m.wikipedia.org/wiki/Probabilistic_latent_semantic_in dexing [5] Blei, David M., Andrew Y. Ng, and Michael I. Jordan. "Latent dirichlet allocation." the Journal of machine Learning research 3 (2003): 993-1022.

[6] Mei, Qiaozhu, Xu Ling, Matthew Wondra, Hang Su, and ChengXiang Zhai. "Topic sentiment mixture: modeling facets and opinions in weblogs." In Proceedings of the 16th international conference on World Wide Web, pp. 171-180. ACM, 2007. [7] Titov, Ivan, and Ryan McDonald. "Modeling online reviews with multi-grain topic models." In Proceedings of the 17th international conference on World Wide Web, pp. 111-120. ACM, 2008. [8] Lu, Yue, ChengXiang Zhai, and Neel Sundaresan. "Rated aspect summarization of short comments." In Proceedings of the 18th international conference on World wide web, pp. 131-140. ACM, 2009.

[9] Brody, Samuel, and Noemie Elhadad. "An unsupervised aspectsentiment model for online reviews." In Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, pp. 804-812. Association for Computational Linguistics, 2010. [10] Jo, Yohan, and Alice H. Oh. "Aspect and sentiment unification model for online review analysis." In Proceedings of the fourth ACM international conference on Web search and data mining, pp. 815-824. ACM, 2011. [11] Lin, Chenghua, and Yulan He. "Joint sentiment/topic model for sentiment analysis." In Proceedings of the 18th ACM conference on Information and knowledge management, pp. 375-384. ACM, 2009. [12] Lin, Chenghua, Yulan He, Richard Everson, and Stefan Rüger. "Weakly supervised joint sentiment-topic detection from text." Knowledge and Data Engineering, IEEE Transactions on 24, no. 6 (2012): 1134-1145. [13] Wang, Hongning, Yue Lu, and ChengXiang Zhai. "Latent aspect rating analysis without aspect keyword supervision." In Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining, pp. 618626. ACM, 2011. [14] Kim, Suin, Jianwen Zhang, Zheng Chen, Alice H. Oh, and Shixia Liu. "A Hierarchical Aspect-Sentiment Model for Online Reviews." In AAAI. 2013. [15] Moghaddam, Samaneh, and Martin Ester. "The flda model for aspect-based opinion mining: addressing the cold start problem." In Proceedings of the 22nd international conference 58

Big Data Analytics on World Wide Web, pp. 909-918. International World Wide Web Conferences Steering Committee, 2013. [16] Xueke, Xu, Cheng Xueqi, Tan Songbo, Liu Yue, and Shen Huawei. "Aspect-level opinion mining of online customer reviews." Communications, China 10, no. 3 (2013): 25-41. [17] Liang, Jiguang, Ping Liu, Jianlong Tan, and Shuo Bai. "Sentiment Classification Based on AS-LDA Model." Procedia Computer Science 31 (2014): 511-516.

[18] Wu, Yao, and Martin Ester. "FLAME: A Probabilistic Model Combining Aspect Based Opinion Mining and Collaborative Filtering." In Proceedings of the Eighth ACM International Conference on Web Search and Data Mining, pp. 199-208. ACM, 2015. [19] Tan, Shulong, Yang Li, Huan Sun, Ziyu Guan, Xifeng Yan, Jiajun Bu, Chun Chen, and Xiaofei He. "Interpreting the public sentiment variations on twitter." IEEE Transactions on Knowledge and Data Engineering 26, no. 5 (2014): 1158-1170.

59

Cognitive Knowledge Engineering

Chapter 13

A Review for Opinion Extraction and Analysis based on Semi-Supervised Learning using Knowledge Base S. C. Nandedkar

J. B. Patil

P. A. Joshi

Dept. of CSE,DIEMS Aurangabad

R.C.Patel COE, Shirpur

Dept. of CSE,DIEMS Aurangabad

sugandhanandedkar@dietms. org

[email protected]

[email protected]

ABSTRACT Internet is becoming user centric and people preferring to exchange opinions through online social means such as ± discussion forums, blogs and micro-blogs. This user opinion EDVH LV D YDOXDEOH UHVRXUFH IURP EX\HUV DV ZHOO DV VHOOHUV¶ perspective. It is helpful for buyers to choose a good product and helpful for sellers to improve their product. The proposed system is intended for opinion extraction and analysis. It makes use of semi-supervised learning approach. Finally it creates the knowledge base to fulfill the said intension.

GENARAL TERMS Unstructured corpus, Preprocessing,

words that is the opinion of the customer is to be analyzed. However, it is necessary to have prior knowledge of opinion word lexicons [2]. Opinion target is defined as the object about which user expresses their views. It is also called as features [4]. For example, features for a mobile handset can be screen, battery, resolution, processing speed, etc. Feature extraction means finding out customer comments related to these features. This is called as Part of Speech (POS) analysis. In this case noun, adjective, verbs and adverbs are the basic form of sensing opinion. As the opinion words usually co-occur with the opinion target, a collective extraction strategy is adapted here. It follows supervised learning approach, i.e. classification with fixed target classes [2, 4, 5, 6].

2. LITERATURE REVIEW KEYWORDS Opinion Mining, Sentiment Analysis, Dependence Analysis, Part of Speech (POS) tagging

1. INTRODUCTION

-

-

Internet is becoming increasingly user centric because of emerging communication platform Web 2.0. People prefer exchanging opinions through online social means such as ± discussion forums, blogs and micro blogs (e.g. face book, Twitter, customer reviews etc.). Along with such trends, huge amount of user generated content are present on the Internet [1]. It consists of rich opinion and sentiment information. We now have a huge volume of this opinion data. Appropriate recognition and analysis of this opinion and sentiment information has become increasingly important and key influencer of our behavior. For service / product providers, LW LV LPSRUWDQW WR NQRZ FXVWRPHUV¶ IHHGEDFN IRU TXDOLW\ improvement. For customer / user of the service, it is important WRNQRZRWKHUV¶IHHGEDFNWRILQGRXWZKLFKLVWKHEHVWSURGXFW service [2]. The process is named as Opinion Mining (OM) or Sentiment Analysis. As given in [3], there is a subtle difference in these two terms. Opinion Mining: It can be defined as sub discipline of FRPSXWDWLRQDO OLQJXLVWLFV WKDW IRFXVHV RQ H[WUDFWLQJ SHRSOH¶V opinion from the Web. Given a piece of text, opinion mining system analyses Which part is opinion expression? Who wrote the opinion? What is being commented? Sentiment Analysis: It is about determining the subjectivity, polarity (positive / negative) and polarity strength (weakly positive, mildly positive, strongly positive) of the piece of text. What is the opinion of the writer? Now, both the customers as well as merchants need a precise RSLQLRQ IURP WKH FXVWRPHU¶V RSLQLRQ 7R GR WKLV ERWK RSLQLRQ target that is the feature of the product or service and opinion

Kang Liu et al. [4] analyzed the relationship between opinion targets and opinion words. This paper proposed a novel approach based on the partially-supervised alignment model, which regards identifying opinion relations as an alignment process. Then a graph-based co-ranking algorithm is exploited to estimate the confidence of each candidate. Finally candidates with higher confidence are extracted as opinion words. D. Ostrowski [5] proposed a methodology for the identification RI WRSLFV DVVRFLDWHG ZLWK FXVWRPHUV¶ VHQWLPHQW XVLQJ D )LVKHU Classification based approach towards sentiment analysis. By considering specific mutual information and word frequency distribution, topics are then identified within sentiment categories. The goal is to provide overall trends in sentiment along with associated subject matter. They demonstrated this methodology against data collected among a particular product line obtained from Twitter advanced search. Saeideh Shahheidari et al. [6] described how to automatically collect Twitter corpus and built a simple sentiment classifier by utilizing the Naive Bayes model to determine the positive and negative sentiment of a tweet. Lastly they tested the classifier DJDLQVW D FROOHFWLRQ RI XVHUV¶ RSLQLRQV IURP ILYH LQWHUHVWLQJ domains of Twitter, i.e. news, finance, job, movies, and sport. Marius Muja et al. [7] proposed a new algorithm for approximate nearest neighbor matching. For matching high dimensional features, they used the randomized k-d forest and a new algorithm proposed in this paper, the priority search kmeans tree. One more new algorithm is proposed for matching binary features by searching multiple hierarchical clustering trees. This paper showed that the optimal nearest neighbor algorithm and its parameters depend on the data set characteristics and described an automated configuration procedure for finding the best algorithm to search a particular data set. In order to scale to very large data sets that would otherwise not fit in the memory of a single machine, they 60

Big Data Analytics proposed a distributed nearest neighbor matching framework that can be used with any of the algorithms described in the paper.

3. PROPOSED SYSTEM The activities involved in the process of opinion mining are: Data Gathering / Crawling, Preprocessing, POS Tagging, Dependence Analysis, Extraction of Feature ± Opinion, Performance Measurement.

3.1 Data Gathering / Crawling As E-commerce has propagated to a vast extend, more people are buying and selling more products online. The customer reviews that describe experiences with product and service use are becoming more important [1]. Potential customers want to know the opinions of existing customers to garner information about the products they plan to buy, and businesses want to find and analyze public or customer opinions of their products to establish future directions for improvement [2]. Customer reviews generally contain the product opinions of many customers expressed in various forms including natural language sentences. Note that the target data set must be large enough to contain different patterns while remaining concise enough to be mined within an acceptable time limit. For feature - opinion extraction WKHFXVWRPHUV¶UHYLHZLVEHVWVXLWDEOe data. This can be crawled from any online shopping website.

3.2 Preprocessing / Data Scrubbing User generated content (UGC) are mostly in unstructured format. UGC is also described as incomplete, noisy and inconsistent [8,9]. It is very difficult to process such unstructured data. It is necessary to apply certain preprocessing steps to the data so that it can be used for further analysis purpose. There are various preprocessing tasks such as, Data cleaning (identify and load in missing values, rectify noisy data, find and mark any outliers if present, and eliminate inconsistencies etc.), Data integration (combining multiple databases, multi dimensional data, or files etc.), Data transformation (normalization and aggregation etc.), Data reduction (decrease the data size or data volume without affecting the analytical results etc.), Data discretization ( it can be considered same as data reduction but replacing numerical values either with nominal ones or bucker of range). For opinion mining and analysis choosing appropriate tool for data preprocessing is essential. It creates a huge impact on the performance of entire system. In the initial step of preprocessing for opinion mining one should eliminates the unnecessary content, such as tags, dates, and reviewer names, from the collected review data. Then, to extract noun phrases from the review data as feature candidates, NLProcessor [13,14] is used to perform morphological analysis, including POS tagging.

3.3 Part of Speech Tagging Part of Speech (POS) tagging is performing the task of dividing the text into words called as token. Then it assigns a tag (identifying that whether it is a noun, adjective, verb etc.) to each token. Stanford POS Tagger can be used for this purpose. This speech tagger is using the model of maximum entropy. It enhances the performance by increasing the information sources used for tagging, incorporating more extensive treatment of capitalization for unfamiliar words. It also addresses the disambiguation of the tense forms of verbs, and focuses more on features for disambiguating part of words from prepositions and adverbs[15].

3.4 Dependence Analysis Once the task of POS tagging is over, the next step is to extract the feature opinion pairs from the tagged tokens. This step is affected a lot by the nature of statement formation done by opinion writer. Few sentences are having clear subject and object part defined whereas few sentences are not having such clear separations [12]. So, it is not possible to extract opinions from all sentences. As a machine learning one can extract direct opinions only. The part of indirect opinion extraction is not considered here. To extract candidate feature ± opinion pairs different combination of dependences are used. The dependency parser finds out the different dependence relations between word pairs. For word w1 and word w2 , the dependence relationship is represented as relation_type(w1 , w2), in which w1 is called lead word and w2 is called dependent or modifier. The relationship relation_type(w1 , w2) can be either direct or indirect.

3.5 Feature Opinion Extraction As mentioned above for opinion extraction nsubj is the most important relationship. The process is feature ± opinion pair extraction uses it. The relationship nsubj (nominal subject) is a noun phrase which is the syntactic subject of a clause. The governor of this relation might not always be a verb: when the verb is a copular verb, the root of the clause is the complement of the copular verb, which can be an adjective or noun[10,11, and 16]. These three cases namely governor as noun, governor as adjective and governor as verb should be treated separately. As per their governor one should change the extraction rules. The extracted information component is represented by a triplet < f , m ,o >, where f stands for a feature generally articulated as a noun phrase, o stands for the opinion word which is generally articulated as adjective, and m is an adverb that acts as a modifier to represent the degree of expressiveness of the opinion. For the present work only the feature ± opinion pairs are considered. First extract the nsubj relationship. In line with above discussion then extract the respective feature from the UHODWLRQVKLS 7KHQ ILQG LW¶V FRUUHVSRQGLQJ RSLQLRQ EDVHG RQ above mentioned rules. Once this step is over now the feature ± opinion base is ready. The further step will find the performance of feature ± opinion extraction process.

3.6 Performance Measurement Evaluation of the experimental results can be performed using standard Information Retrieval (IR) metrics of Precision and Recall, these are defined in the below equations-

In these equations, TP indicates true positive, which is defined as the number of feature-opinion pairs that the system identifies correctly, FP indicates false positive which is defined as the number of feature-opinion pairs that are identified falsely by the system, and FN indicates false negatives which is the number of feature-opinion pairs that the system fails to identify. Based on the above criteria one can check for the performance of the system.

4. CONCLUSION: The proposed system works for feature ± opinion extraction purpose. For small data set of customers review it shows proper results. Considering the UGC feature it gives less accuracy. 61

Cognitive Knowledge Engineering To increase the performance of the system one should pay more DWWHQWLRQ RQ WKH QDWXUH RI XVHU¶V ZULWLQJ VW\OH DV ZHOO DV WKH

5. REFERENCES [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

W. Xindong, Z. Xingquan, W. Gong-4LQJDQG:'LQJ³'DWD Mining with Big 'DWD´ IEEE Trans. Knowledge and Data Engineering, pp. 215-227, July 2013. % /LX ³6HQWLPHQW $QDO\VLV DQG 2SLQLRQ 0LQLQJ´ Synthesis Lectures on Human Language Technologies, vol. 5, no. 1, pp. 1167, May 2012. $ 'DV ³2SLQLRQ ([WUDFWLRQ DQG 6XPPDUL]DWLRQ from Text 'RFXPHQWV LQ %HQJDOL´ 3K ' GLVVHUWDWLRQ -DGDYSXU University, Department of Computer Science & Engineering, Kolkata, December 2011 ./LX;/LKHQJDQG-=KDR³&R-extracting Opinion Targets and OpinionWords from Online Reviews Based on the Word $OLJQPHQW 0RGHO´ IEEE Trans. Knowledge and Data Engineering, vol. 6, no. 1, January 2013. ' 2VWURZVNL ³6HQWLPHQW 0LQLQJ ZLWKLQ 6RFLDO 0HGLD IRU 7RSLF ,GHQWLILFDWLRQ´ LQ Proc. IEEE Fourth International Conference on Semantic Computing, pp. 394 ± 401, 2010. 6 6KDKKHLGDUL + 'RQJ DQG 0 1 5 'DXG ³Twitter VHQWLPHQW PLQLQJ $ PXOWL GRPDLQ DQDO\VLV´ LQ Proc. IEEE Seventh International Conference on Complex, Intelligent, and Software Intensive Systems, pp. 144 ± 149, 2013. 00XMDDQG'/RZH³6calable Nearest Neighbor Algorithms IRU+LJK'LPHQVLRQDO'DWD´IEEE Trans. Pattern Analysis and Machine Intelligence, pp. 2227 ± 2240, vol. 36, no. 11, Nov.2014. A. Tatu, L. Zhang, E. Bertini, T. Schreck, D. Keim, S. Bremm, DQG7/DQGHVEHUJHU³&OXVW1DLOs: Visual Analysis of Subspace

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

different dependence relationships generated by dependence parser. &OXVWHUV´Tsinghua Science and Technology, ISSN l1007-0214 1l05/11, vol. 17, no. 4, Aug. 2012, pp. 419 ± 428. 1 3UHQGLQJHU DQG 0 ,VKL]XND ³6HQWL)XO *HQHUDWLQJ D 5HOLDEOH /H[LFRQ IRU 6HQWLPHQW $QDO\VLV´ ,((( 7UDQV Affective Computing, vol. 2, no.1, pp. 22 ± 36, June 2011. 3%DODPXUDOL'0DQQDDQG3%KDWWDFKDU\\D³&URVV-Domain Sentiment Tagging Using Meta Classifier and a High Accuracy In-'RPDLQ &ODVVLILHU´ LQ SURF RI (LJWK ,QWHUQDWLRQDO Conference on Natural Language Processing ICON 2010. V. Subrahmanian and D. Reforgiato, ³$9$ $GMHFWLYH-VerbAdverb CombinDWLRQV IRU 6HQWLPHQW $QDO\VLV´ ,((( 7UDQV Intelligent Systems, vol. 23, no. 4, pp. 43 ± 50, July 2008. )3HOHMD-6DQWRVDQG-0DJDOKDHV³5DQNLQJ/LQNHG-Entities LQ D 6HQWLPHQW *UDSK´ LQ 3URF ,QW MRLQW &RQI :HE Intelligence and Intelligent Agent Technology, IEEE/WIC/ACM, pp. 118-125, 2014. S. Tan, Y. Li, H. Sun, Z. Guan, X. Yan, J. Bu, C. Chen, and X. +H ³,QWHUSUHWLQJ WKH 3XEOLF 6HQWLPHQW 9DULDWLRQV RQ 7ZLWWHU´ IEEE Trans. on Knowledge and Data Engineering, vol. 6, no. 1, September 2012. B. Wang, Z. Xiao, Y. Liu, and Y. Xu³6HQWL9LHZ6HQWLPHQW $QDO\VLV DQG 9LVXDOL]DWLRQ IRU ,QWHUQHW 3RSXODU 7RSLFV´ ,((( Trans. Human Machine Systems, pp. 620 ± 630, Oct. 2013. 0 1DJDL 0 2QR DQG 5 6KLEDVDNL ³,QWHURSHUDELOLW\ IRU *OREDO 2EVHUYDWLRQ 'DWD E\ 2QWRORJ\ ,QIRUPDWLRQ´ 7LQJKXD Science and Technology, ISSN 1007-0214 54/67 pp. 336-342, 2008. Z. Hai, K. Chang, J. Kim, DQG&= Threshold(th). 5. Slide over the Window. 6. Calculate the transitive closure

Fig 1 : Comparison space comparison of DCS ,DCS++ and SNM for Cora Dataset

Further Enhancements are made to this basic algorithm with DCS ++ algorithm. In the DCS++ (w-1) records are added. The transitive property is used to minimize the number of comparisons. DCS++ saves (w-2) comparisons per duplicate.

2. EXPERIMENT ON CORA DATASET We have used Cora (real) dataset for experiment. It has1296 records and 116 true duplicates. Duplicate distribution is uneven in the dataset. We have conducted experiment for SNM ,DCS and DCS++ algorithms. We have taken Cora dataset and set the threshold to 80 %for similarity measure. The token is author. Fig.1 shows the graph of comparisons required for SNM, DCS and DCS++ algorithms. We can see that DCS++ needs less than 50 % comparisons than SNM. The Fig 2. Shows the Pair completeness (PC) and Reduction Ratio(RR) values for all three algorithms. It shows that PC and RR values of SNM are better than DCS and DCS++ algorithm. PC evaluates the quality of blocking scheme . It evaluates the number of true duplicates . We can see that SNM performs better in the Quality for determining true duplicates though it needs more comparisons than DCS and DCS++.

Fig 2 : Comparison for Pair completeness and Reduction ratio of DCS ,DCS++ and SNM for Cora Dataset

3. EXPERIMENT ON RESTAURANT DATASET Restaurant contains 864 restaurant names and addresses with 112 duplicates. It has at most two duplicates in a window. There are lesser number of erroneous records in Restaurant dataset. . The blocking key we use is the first 4 characters of the restaurant name concatenated with the first 4 characters of the city name.

RR is a measure of blocking scheme . It evaluates the number of record pairs to be compared in detail in the candidate set. High Reduction ratio suggests that the blocking technique has considerably removed many pairs from the full comparison space. The low reduction ratio means that a larger number of candidate record pairs have been generated. In the fig. 2 RR shows low values for SNM while PC values are is better than DCS and DCS++.

Fig 3 : Comparison space comparison of DCS ,DCS++ and SNM for Restaurant Dataset

We can see that DCS , DCS++ and SNM have nearly equal number of comparisons . DCS and DCS++ did not out perform in this dataset as the duplicates are at most two per window. So in this case window 97

Cognitive Knowledge Engineering it gives better reduction in comparison for Cora dataset where the cluster sizes vary a lot. But it did not successful for restaurant dataset where the cluster size is almost same i.e. two for every cluster. Thus we can conclude it is useful only for varied cluster sized dataset.

sizes doses not vary a lot. So the SNM and DCS algorithms give equal performance in such datasets.

4. CONCLUSION SNM is a standard algorithm for duplicate detection. As it is based on fixed sized windowing method it may ignore the duplicates if the size of the window is smaller than the number of possible duplicates. It also suffers from unnecessary comparisons if the size of the window is bigger than the number of possible duplicates. DCS and DCS++ are the algorithms uses the optimum varying window size to detect duplicates. It uses transitive closure for reducing number of comparisons. Our experiment shows that DCS and DCS++ algorithms reduce the comparison space to reduce the complexity. We also found that

5. REFERENCES [1]

[2]

0 $ +HUQDQGH] DQG 6 - 6WROIR ³7KH PHUJHSXUJH SUREOHP IRU ODUJH  GDWDEDVHV´ LQ 3URceedings of the ACM International Conference on Management of Data (SIGMOD), 1995, pp. 127±138. 0 $ +HUQDQGH] DQG 6 - 6WROIR ³5HDO-world data is GLUW\ 'DWD  FOHDQVLQJ DQG WKH PHUJHSXUJH SUREOHP´ 'DWD Mining and Knowledge Discovery, vol. 2(1), pp. 9±37, 1998.

Also it is found that it does not contribute much to the quality of the duplicate detection. We have seen that SNM outperforms DCS algorithms in true positive coverage. Thus DCS will be alternative to SNM for large datasets where comparison space matters a lot for complexity issue. There is always a tradeoff between the comparison space and quality of true positive coverage.

[3]

: , &KDQJ DQG 7 * 0DUU ³$SSUR[LPDWH VWULQJ PDWFKLQJDQGORFDOVLPLODULW\³&RPELQDWRULDO3DWWHUQ0DWFKLQJ Lecture Notes in Computer Science, pp. 259±273, 1994.

[4]

U. Draisbach, F. Naumann, S. Szott, and O. Wonneberg, ³$GDSWLYH :LQGRZV IRU 'XSOLFDWH 'HWHFWLRQ´ ,((( WK International Conference on Data Engineering, 2012.

[5]

0 %LOHQNR DQG 5 - 0RRQH\ ³$GDSWLYH GXSOLFDWH GHWHFWLRQ XVLQJ OHDUQDEOH VWULQJ VLPLODULW\ PHDVXUHV´ LQ Proceedings of the International Conference on Knowledge Discovery and Data Mining (KDD), 2003, pp. 39±48.

98

Human Computer Interaction

Human Computer Interaction

99

Cognitive Knowledge Engineering

100

Human Computer Interaction

Chapter 21

A Survey on Vision Based Static and Dynamic Hand Gesture Systems for Mobile Devices A.V.Dehankar

Sanjeev Jain

Department of Computer Technology Priyadarshini College of Engineering,Nagpur

Shri Mata Vaishno Devi University, Jammu

V. M. Thakare Dept. of CSE,Amravati University,Amravati

[email protected]

archana_dehankar@rediff mail.com ABSTRACT Since the inception of computers, application programs captures input through hardware devices like keyboard, mouse, mike, joystick, pen etc. Similarly for mobile devices, the input is given through keypads, mike, touch etc and most recent mobile devices make use of touch screen, voice recognition, geographical location etc. to capture the input. On the other hand, rate of introducing new computing software and applications in market is at higher side as compared to user interface applications. Humans are more inclined towards easy to use interface when interacting with computers or mobile devices. Verbal and non verbal communication is equally important in human life and for interacting with technical devices of this modern era like computers, mobiles etc. Non verbal communication methods such as sign language, gesture and posture are in existence since ancient times. These methods of interaction are used to develop userfriendly interfaces for communicating with computers, mobiles, machines etc. In a Gesture based communication system, motion of hands or body is used to communicate the message. Many researchers are using hand gestures to develop user interface through which computers, mobiles etc. are controlled and handled in an effective manner. The important steps to implement a gesture based user interface include acquiring the gesture, processing / segmenting it to correctly recognize the gesture for completing the desired action. In this paper, we highlight the current research carried out in implementation of user interfaces using hand and finger based gesture recognition systems for mobile devices, we present the different methods and their recognition accuracy and detailed study analysis of implemented systems with various parameters.

meaningful body motions that are movements of hands, arms or fingers. Hand gesture is either static or dynamic. A static hand gesture constitutes stable hand or finger shape whereas dynamic gesture is made up series of movements of hand or fingers. Dynamic gesture recognition is quite complex but it is more practical as compared to static gesture recognition [1,6,7,8]

1.1 Glove Based Hand Gesture Recognition In a Glove Based Hand Gesture Recognition System, a user wears ³'DWD *ORYH´ HPEHGGHG ZLWK ZLUHG WDFWLOH RU VHQVRU\ XQLWV attached to the fingers or joints of the glove and it uses wired or wireless communication path to host device like computer. The sensory unit contains mechanical or optical sensors that estimates the movements of joints and fingers which is mapped with unique hand gestures that computer interprets to take desired actions. The data gloves are of two types :Active and Passive. The active data gloves measures the flexures or acceleration of hand or fingers and they establish communication with host device through wired or wireless medium. The passive data gloves use external devices like camera to recognize the figure gesture using colors or markers instead of sensors [1]. The data glove based systems involve heavily hardware parts such as mechanical or optical sensors embedded in the glove. Normally acoustic or magnetic sensors are used to transmit finger flexures into electrical signals to determine the hand postures [2].

1.2 Drawing Gestures or Touch Based Systems

Keywords

The Drawing Gesture or Touch based systems use stylus or input is given through the touch and patterns made using hand finger tips. Such systems use sensors / virtual environment for recognizing the drawn gesture or patterns. These systems use mechanical sensing because of which accuracy, reliability and electromagnetic noise are some of the problems associated with these systems. These systems require external hardware for implementation [2].

Vision based systems, static and dynamic hand gestures, Segmentation & Tracking, Feature extraction

1.3 Vision Based Systems

General Terms Human Computer Interaction

1. INTRODUCTION AND HISTORICAL BACKGROUND OF HAND GESTURE BASED SYSTEMS In early days humans invented sign language based on hand gestures to communicate with each other. Human hand gestures provide the natural and effective mode of non-verbal communication with the computer Interface. Hand gestures are the

Vision based systems are thought to be very effective and gaining attention of researchers. The base of such system is the way in which individual person receives and react on the information received through movements of body parts such as head, nose, face, fingers, legs etc. Utilizing gesture visuals to recognize the meaning and perform required action becomes more practical with the use of visual sensing and it does not require any additional hardware to wear on hands. To build a vision based hand gesture system only a camera, webcam, camcorder or anything that 101

Cognitive Knowledge Engineering captures the image and communicates with computer or mobile device is required [2]. In this paper, our focus is on visual based hand gesture recognition systems. Hence, in next section we discuss the related work on vision based hand gesture recognition systems developed for mobile devices.

2. RELATED WORK As we are more interested in less hardware involved in hand gesture recognition systems utilized to design an user interface, our main focus is on studying and finding the techniques that are more suitable, accurate and provide faster performance in operation and controlling of functions / applications in smart phones / mobile devices. In this section, we have studied vision based hand gesture recognition systems and summarized the systems that employ various techniques at each level of building a hand gesture recognition system for implementation of user interface or similar applications. In literature, we found several of methods and techniques that are used to build user interfaces for mobile devices and computers. In this section firstly, we discuss the most recent work carried out in vision based static and dynamic hand gesture recognition systems for mobile devices and summarize the analysis of methods used to implement such systems.

2.1 Vision Based Static Hand Gesture Systems for Mobile Devices In this section, we discuss the most recent work carried out on vision based static hand gesture recognition systems for mobile devices available in literature. Tejashri J. Joshi and et al., have implemented a new approach to recognize hand gesture system for static gestures captured through 5.0 MP RGB Camera of android device with processor 1.2 GHz and having 1GB RAM. The system works well in constrained lighting environment with a black background with five gestures RI ILQJHUV IURP RQH WR ILYH WUDLQHG DQG WHVWHG IRU  SHRSOH¶V gestures and provided 97.6% accuracy for the combination of PCA & Decision tree based multiclass SVM recognition technique. The authors also tested combination of PCA with Euclidian Distance and K-means clustering for which they got accuracy 72.4% and 45% respectively. The approach contains three steps: preprocessing, feature extraction and classification. The input image is obtained on device through camera at 10 frames per second and segmentation is carried out using thresholding method. PCA is used to extract the features from region of interest and decision tree based multiclass SVM is used to recognize the static gesture. All the SVM models at each node of the tree are stored in the XML files and system is implemented in Android using OpenCV library. The authors also suggested that further work should focus on hand segmentation method on the resource constrained devices with varying light conditions and skin color background. As we increase the gesture set, the tree based SVM classifier will increase the computation cost. So to minimize the computational cost of the tree based SVM, one may include specialized multiclass SVM classifier [3]. Jie Song and et al., have presented a random forest based real time gesture recognizer that can run on mobile devices, including tablets, Smartphone and smart watches. The algorithm takes input images from a regular RGB camera and does not require any other hardware modification. The authors also proposed a number of compelling interaction scenarios, allowing users to seamlessly transition between touch and gestural interaction, and allowing for

bi-manual, simultaneous touch and gesture interaction. To achieve maximum accuracy for minimal memory footprint authors introduced multi-stage classification forests [4]. The algorithm robustly recognizes wide range of in-air gestures, supporting user variation, and varying lighting conditions. They demonstrated numerous interaction tasks such as mode switches, application and task management, menu selection and certain types of navigation, where such input can be either complemented or better served by in-air gestures. This removes screen real-estate issues on small touch screens, and allows input to be expanded to the 3D space around the device. They presented results for recognition accuracy (93% test and 98% train), impact of memory footprint and other model parameters. For segmentation and preprocessing, the authors used skin color detection thresholding method which is easy to implement, computationally cheap and provides good compromise between true and false values. A hand state classification method makes use of shape i.e. binary masks to infer hand states and features [4]. In this section, we discussed the most recent work carried out on vision based static hand gesture recognition systems for mobile devices available in literature.

2.2 Vision Based Dynamic Hand Gesture Systems for Mobile Devices In this section, we discuss the most recent work carried out on vision based dynamic hand gesture recognition systems for mobile devices available in literature. M. Favorskaya, A. Nosov, and A. Popov have proposed method for dynamic gesture recognition that included variant time trajectory classifiers and the posture classifiers that extracted subgestures in selected time instants using a rule based method of recognition. The software tool is developed in C# language using visual studio 2012 framework and it is named as DynGesture and implements two main algorithms. In segmentation algorithm, localization of hand is done through skin color classifier for building a topological skeleton from palm and fingers so that a trajectory classifier and sub gesture description vector can be implemented. In recognition algorithm, five rules are used to calculate positions of fingers from each sub gesture called posture classifier. Then decision tree procedure is used to analyze the descriptor vector of sub gesture sequence. In this system Microsoft Kinect Camera is used to capture the images in uniform background. The authors used 393 gestures against approximately 13858 images dataset; the gestures are recognized by detecting the fingertips and achieved 84-91% accuracy in a uniform background. The system is applicable for learning and recognizing gestures of sign language [5]. Jun-ho An and etl., addressed problem of two handed interaction is required with front facing camera and proposed a method that can be handled through single hand which holds mobile phone with the camera and makes use of rear facing camera to capture the gestures. They used various finger tip gesture to perform Click and move operations up, down, left & right. The authors used Skin Color Segmentation, Morphological operation and skeletonization for finger tracking. This system recognizes five dynamic gestures in varying background with uniform and low lighting conditions using 1.3 MP Camera. The system is applicable to generate mouse click events and commands through gestures. The gesture recognition accuracy is LQXQLIRUPOLJKWLQJFRQGLWLRQVZKHQWKHEDFNJURXQGGRHVQ¶W contain color similar to skin color [6]. 102

Human Computer Interaction Orazio Gallo and et al. has proposed a vision based pointing system that allows the users to control the pointers position by just waving a hand, with no need for additional hardware. The system recognizes three dynamic gestures using finger tips in uncluttered background through QVGA camera at 30 fps on Nokia N95 mobile phone and tested with 13 images. The system is applicable to control pointer position and user can easily drag the image to explore various regions in an image viewer application through gestures. The Vision based pointing system that allows the user to control pointers position by just waving a hand with no need for additional hardware is implemented in Symbian C++ using fixed point arithmetic and interface and camera is controlled through Python for S60. The system addressed Factors like camera shake,

poor picture resolution and the algorithm takes 20 ms to execute [7].

3. STUDY ANALYSIS In this section the tables below describes the methods that are currently used to implement vision based hand gesture recognition systems and their system specification with various parameters. In table 1 various Methods used during the different steps or stages to implement vision based static and dynamic hand gesture recognition systems that are available in literature. During the study analysis ,we have considered important system parameters include gesture type, Camera & FPS, Background, Lighting Conditions, No of gestures, No of Images / training set, recognition accuracy and possible applications / contributions etc. Our findings are revealed in conclusion section.

Table1: Methods and Recognition Accuracy

TABLE 2. Analysis of Implemented Systems with Various System Parameters

Method [Ref No.] [3]

Gesture Type

Camera & FPS

Background

Lighting Condition

No of Gestures

Static

5.0 MP RGB Camera 10 fps

Black background

Constrained

05

No of Images / Training Set 35

Key points / Recognition through

Recognition Accuracy

Possible Applications / Contributions

Fingers

97.6%

A new approach to recognize the gesture.

103

Cognitive Knowledge Engineering [4]

Static

RGB camera

Uniform background (black)

varying lighting conditions

07

2D Images 58,000

Finger tips and pinch

93% test and 98% train

tasks such as mode switches, application and task management, menu selection and certain types of navigation

[5]

Dynamic

[6]

Dynamic

[7]

Dynamic

Microsoft Kinect camera 1.3 MP Camera

Uniform background

--

393

13, 858

Finger tips

84-91%

Varying background

Uniform & low lighting

05

--

Finger tips

88% In uniform lighting

QVGA 30 fps

Uncluttered Background

--

3

13

Finger tips

Learning system of sign language Mouse events and commands To control pointers position. User can drag the image to visit different regions.

4. CONCLUSION Our study reveals that to implement vision based static and dynamic hand gesture recognition systems useful methods at various stages like preprocessing or segmentation includes thresholding, and skin color based segmentation, morphological operations and classifiers build around the skin color. The feature extraction techniques include Principal Component Analysis (PCA), Binary Masks created on the basis of shape; Poster Classifiers to extract sub gesture and Skeletonization. The classification or recognition techniques include Euclidian Distance, K-means clustering, Decision Tree Based Multiclass SVM, Multistage classification forests, Rule-based recognition, Decision tree procedure and recognition algorithm. The static hand gesture systems exploits RGB Camera, black background with constrained and varying lightning conditions to recognize fingers, finger tips and or finger pinch providing the recognition accuracy range from 45% to 98 % in combination of various recognition methods. Dynamic hand gesture recognition systems are implemented using cameras like, Microsoft Kinect, QVGA camera for capturing images in uniform and varying background under uniform and low lighting conditions makes use of finger tips as a recognition object and provides recognition accuracy range from between 84% to 91 %. The possible applications of such systems include task management, navigation, switching between modes, learning system for sign language, generating mouse events and command, controlling pointers positions, zoom in and zoom out on images, drag and drop etc. generating mouse events >@  0 )DYRUVND\D $ 1RVRY $ 3RSRY  ³ /RFDOL]DWLRQ DQG recognition of Dynamic Hand Gestures Based on Hierarchy RI 0DQLIROG &ODVVLILHUV ³ 7KH ,QWHUQDWLRQDO $UFKLYHV RI Photogrammetry,Remote sensing and Spatial Information Sciences, Volume XL-5/W6,2015

and command, controlling pointers positions, zoom in and zoom out on images, drag and drop etc.

5. ACKNOWLEDGMENTS Our thanks to the experts who have contributed towards development of the paper

6. REFERENCES [1] P. Premaratne, Human Computer Interaction Using Hand Gestures, Cognitive Science and Technology, Chapter 2 Historical Development of Hand Gesture Recognition Page No. 5- 14, DOI 10.1007/978-981-4585-69-9_2, © Springer Science+Business Media Singapore 2014. [2] http://matlabsproj.blogspot.in/2012/06/hand-gesturerecognition-using-neural.html [accessed on 12/04/2016] [3] Tejashri J. Joshi Shiva Kumar, N. Z. Tarapore, VivekMohile ³6WDWLF+DQG*HVWXUH5HFRJQLWLRQXVLQJDQ$QGURLG Device , International Journal of Computer Applications (0975 ± 8887),Volume 120 ± No.21, June 2015. [4] Jie Song, Gabor Soros, Fabrizio Pece, Sean Fanello, Shahram ,]DGL&HP.HVNLQ2WPDU+LOOLJHV³,Q-air Gestures Around 8QPRGLILHG 0RELOH 'HYLFHV´ 8,67  2FWREHU ±8, 2014, Honolulu ,USA [6]

Jun-KR $Q  .ZDQJ 6ZHRN +RQJ  ³)LQJHU *HVWXUH 0RELOH User Interface Using D 5HDU IDFLQJ FDPHUD´,((( international Conference on Consumer Electronics(ICCE) 2011,pp 303-304.

104

>@  2UD]LR *DOOR 6RQLD 0 $UWHDJD -DPHV ( 'DYLV ³$ FDPHUD EDVHGSRLQWLQJLQWHUIDFHIRUPRELOHGHYLFHV´,&,3,((( pp.1420-1423 [8] Jayesh S. Sonkusare, Nilkanth. B. Chopade, Ravindra Sor & 6XQLO/ 7DGH ³$ 5HYLHZ RQ +DQG *HVWXUH 5HFRJQLWLRQ 6\VWHP´ ,QWHUQDWLRQDO &RQIHUHQFH RQ &RPSXWLQJ Communication Control and Automation

2015 IEEE, DOI 10.1109/ICCUBEA.2015.158 ,pp.790794 [9] ArpLWD 5D\ 6DUNDU * 6DQ\DO 6 0DMXPGHU³ +DQG *HVWXUH 5HFRJQLWLRQ 6\VWHPV $ 6XUYH\´ ,QWHUQDWLRQDO -RXUQDO RI Computer Applications (0975 8887) Volume 71± No.15, May 2013. Page No. 26-37, IJCATM: www.ijcaonline.org

105

Cognitive Knowledge Engineering

Chapter 22

A Review of Software Effort and Cost Estimation using Soft Computing Azade Sanjay Y.

Lomte S. S.

Aqueel Ahmed A. Jalil

Dept. Of C. S. & I. T. Dr. B. A. M. U. Aurangabad

Dr. Seema Quadri Institute of Eng. & Tech. Aurangabad

Department of Computer Science Radhai Mahavidyalaya

[email protected]

[email protected]

ABSTRACT Software effort and cost estimation are more important than anything else, for the success of software development projects. Number or researchers proposed different modeling techniques but it is observed that no one is applicable for all, because every technique is having its advantages and disadvantages. In this paper we have presented software effort estimation models used in early stage of software when most of the things are not clear. We have covered almost all available well known theoretical approaches based on expert judgments and the evolution of soft computing is discussed in the paper and try to comprehend the pros and cones, similarities and differences among these models.

Keywords Software cost estimation models; Effort prediction; software metrics; lines of code, fuzzy logic, FLC, GA.

1. INTRODUCTION ,Q WRGD\¶V PRGHUQ ZRUG WKH GHSHQGHQF\ RQ VRIWZDUH LQFUHDVLQJ continuously almost all products and services are only possible through software system, because of this the size and complexity of software system has increased rapidly and this increased size and complexity in software system has involves as an essential part of fundamental shift in their cost, time to market, functionality and quality requirements [1]. Software cost estimation is used basically to get an approximation of the essential resources needed by a particular software project and their schedules [2]. This review focused on understanding the concept of software metrics, software cost estimation, and evaluation of soft computing software development effort estimation models. Process of software estimation broadly includes four steps. At first we estimate software size, then the needed effort after this we derive the schedule and at last calculate cost of the software. software development is getting complex because of many reasons some of them are , development technology and paradigms change rapidly , development distribution increases , software development is still a largely human intensive process, software products has an abstract character. The complex and multi dependent character of software make its management a challenging task, there for we must focuses on financial success. A good modern estimation method provide planning and negotiation related to cost [6]. Objective of effort and cost estimation are managing and reducing project risk , process improvement and organizational learning ,

ansari_aqueelahmed@rediffm ail.com

base lining and benchmarking productivity, negotiating project resources and scope, managing project changes and reducing management overhead. The structure of this paper is as follows: in section 2 includes classification of effort estimation methods, section 3 includes effort estimation techniques using soft computing, section 4 we have presented future work, in section 5 we have concluded our discussion. 2. CLASSIFICATION OF EFFORT

ESTIMATION METHODS There are number of effort and cost estimation methods invented by software researchers and every newly invented method is found overcoming the shortcomings of previous methods, then also most of the methods are now beyond research and academic applications. Following are the methods used for estimation, among them only few are the known and applied in software development practices. The below effort estimation methods are grouped on the aspects as the type of data used by them as input and the estimation principal the use. This section is an effort to review most of the famous estimation meWKRGVZKLFKPD\SURYLGHUHVHDUFK¶VWRFRPSDUHWKHLUSURVDQG cons or judgment for the selection of particular method for estimation.

2.1 Data driven estimation Refers to methods that predict effort based on solely on the quantitative analysis of historical project data. The relationships found are then projected on to a new project in order to predict the H[SHFWHGHIIRUW'DWDGULYHQHVWLPDWLRQLVRIWZRW\SHV¶SURSULHWDU\ and non-proprietary [7].

2.1.1 Proprietary effort estimation method Refer to methods that are not fully documented in the public domain ,but basic principles and estimation algorithms are often publically available, but the details of estimation procedure are not available to public , these methods are data riven. The advantage of these models is that one can estimate effort without having any historical project data. Project effort is estimated based upon actual effort data collected across similar projects from external organizations. All the project data are hidden in a proprietary software tool. The disadvantage of this 106

Human Computer Interaction method is that the estimate depends on existence of similar project in the undergoing benchmarking data repository.

2.1.2 Nonproprietary models Models are of mainly of three types Model based, Memory based, Composite There are two methods using which effort are predicted a) Explicit and b) Implicit. 2.1.2.1 In explicit estimation that is model based which relates the development effort to one or more project characteristics is developed based on analysis of historical data. 2.1.2.2 In Implicit estimation that is memory based in this approach it is assumed that projects of similar type characteristics require similar effort to be completed so this approach focus on modeling similarity between projects.

2.1.3 Composite methods These Models integrate elements of model based and memory based models. They are used to overcome any weakness of individual technique. Model based estimation use statistical regression and use data to build a parameterized effort model where model parameters might be specified a prior or might be completely learned from data. After training one or more components model with certain modeling techniques is created for prediction and data are not needed at the time of prediction. Model based techniques use quantitative project data to build parameterized effort model. Different approaches are required for determining the models parameters depending on the type of underlying model. Parametric model requires specifying a priori the parameters of the In nonparametric estimation estimator produce their inference free from any particular underlying functional form. In this the number and nature of parameters are flexible and not fixed in advance. Semi parametric estimation contains both parametric and nonparametric components.

2.1.4 Memory based estimation: In memory based estimation it retain the available project data and use them each time a prediction of a new project is to be made. Each time when estimation is to be provided for a new project the previous project data is searched using one or more analogy techniques to find the project that are most similar to the project which is to be estimated. Once appropriate project is found that project is used as an input to estimation techniques. The example of this is case-based reasoning (CBR)[8]. Although model based and memory based techniques share many characteristics common to data driven estimation but they are differ in number of detailed aspects which help in deciding whether a given approach is or not useful in particular estimation context. Advantages of data driven methods: - Minimal involvement of human experts, High flexibility of estimation method, large documentations and tool support, applicability for any project activities, availability at any stage of software development significant amount of empirical evidence and relatively low application cost. Disadvantages of data driven methods extensive requirements regarding quantitative data, relatively high complexity of estimation method, limited reusability of estimation outputs, inconsistence robustness against messy data, little support for handling information uncertainty, inconsistence predictive power

2.2 Expert Based Methods

In these methods estimations are made by experts without using formal estimation models. It involves consulting with one or more experts who use their domain experience as well as understanding of the organization context for estimating project cost. It uses two approaches as single user expert approach and multiuser expert approach [9]. In single user expert approach only one user provides final estimates. In doing so the expert could guess the predicted effort directly (rule of thumb) or estimates on basis of more structured reasoning. In multi user approach a group is invited, the advantage of this approach is that individual estimates which are forgotten in single user approach that also covered by other experts i.e. estimation of some can be balanced by the caution of other user? In these a simple mechanical integration applied commonly in the context of expert based estimation is to capture the mean or median estimates of all individual estimates. Advantages of expert based estimates are few requirements regarding quantitative data, high flexibility of estimation method and simplicity of estimation method. Disadvantages of expert based estimates are extensive involvement of human expert, relatively low robustness against messy inputs, relatively little support, limited reusability of estimation output, inconsistent predictive power, little informative power and moderate support for handling information uncertainty.

2.3 Hybrid methods It is observed that the combination of data driven and expert based estimation approaches can substantially improve the accuracy of estimation. The hybrid method offers the combination of strengths of both strategies while avoiding their weaknesses. The advantages of this method are moderate expert involvement, moderate data requirements, high flexibility, moderate support level, and high reusability, and predictive power, informative power, handling uncertainty, estimation scope and empirical evidence. The disadvantages of these are little robustness and moderate to high complexity of estimation method.

3. EFFORT ESTIMATION TECHNIQUES USING SOFT COMPUTING Soft computing plays very important role in developing software effort estimation techniques, soft computing techniques consist of fuzzy logic system, neural network model and genetic algorithm techniques [11]. The COCOMO is one of the most important model for software cost estimation in [12] soft computing GA technique is used for tuning the parameters of COCOMO for prediction of more accuracy, the performance of this method is compared with NASA dataset and it is observed that the tuning coefficient using GA produce results more accurate which in turn help in administer the resources. Fuzzy logic can easily solve problems that are too hard to be understood, in this FLC using triangular membership function for fuzzy set is provided and conclude that study on NASA 93 datasheet based on the various criteria VFA, MARE, VARE, Mean BRE, MMRE and Pred, the developed fuzzy model using triangular membership function provide good results but the limitation with this structure is that as the number of inputs increases the rule base also increases but only few rules would be fired for simple input all FLC rules have to checked, which increases the response time so rules must be minimized [19]. Fuzzy logic is an easy way for mapping input space to output using certain rules so it has been effectively used for estimation of software projects. In practice several authors proposed FLC (fuzzy logic controller) for software effort estimation [13, 14, 15, 16, 17, and 18]. [20] in this paper fuzzy sets rather than classical intervals in intermediate COCOMO 107

Cognitive Knowledge Engineering proposed by Gaussian membership function in fuzzy technique is used for effort estimation after considering the results attained by means of applying COCOMO mode and GFA models, it is observed that effort estimated by fuzzifying size and effort using 13 GMF yields batter results and very near to actual results and it is concluded that the increasing the number of membership functions the performance of FIS improved. [21] In the early stages of the software development life cycle. Software effort estimation model based on fuzzy system can overcome characteristics of uncertainty exist in software effort drivers. However, the determination of the suitable fuzzy rule sets for fuzzy inference plays an important role in coming up with accurate and reliable effort estimates. Software effort estimation based on fuzzy logic is an attempt in the area of software project estimation. The objective work is to provide a technique for software cost estimation that performs better than other techniques on the accuracy of effort estimation. The work has shown by applying fuzzy logic on the algorithmic and non-algorithmic software effort estimation models accurate estimation is achievable

4. FUTURE WORK We have tried to cover most of the software effort estimation approaches and effort estimation techniques using soft computing, but still few are discovered during research but not discussed in this paper. In future research could be conducted about all models which are left in this paper. Cost estimation become more reliable and sophisticated if some new technique using soft computing and algorithmic / non- algorithmic techniques in cost estimation is merge with some new approaches like Genetic algorithm further study can perform to check this assumption.

5. CONCLUSIONS AND SUGGESTIONS In this paper we have summarized a number of estimation approaches and models other than few techniques in nonalgorithmic section. The paper started with a concise overview of understanding cost estimation and cost estimation methods / approaches in section 2 of this paper we have discussed these approaches in detail with advantages and disadvantages of each. In section 3 we have discussed effort estimation techniques using soft computing and we observed that all researchers provide something new and valuable for estimation, but then also ever approach is not without problems. Every technique propose to solve the problem while working in its own environmental boundaries and apparently the algorithms developed at one environment may not possible to be utilized at other environment but we have seen that COCOMO II with soft computing is an enhanced approach then others and the advancement of COCOMO II with soft computing approach though not discussed, is continuously evolving. During review it is observed that most of the researchers have conducted their experimental work on COCOMO II that is why it might be better using COCOMO II and soft computing and put some other batter version of COCOMO II parametric model in this race. It is observed that any alone approach either algorithmic using mathematical proven formula or expert based technique is not sufficient, but if these techniques merged to estimate effort and other factors expert based techniques can be used to estimate and can be validated by using parametric model technique. And might be possibility the combinational technique using soft computing gives best result

6. REFERENCES [1] Software project effort estimation foundations and best practice guidelines for success Adam Trendowicz Ross, Jeffery Springer ISBN 978-3-319-03628-8(eBook) >@ 3UHVVPDQ 5RJHU 6´ 6RIWZDUH (QJLQHHULQJ $ 3UDFWLWLRQHU¶V $SSURDFK´WK (GLWLRQ 0F*UDZ- Hill, New York, USA,ISBN :13:9780073019338,2005. [3] Software Development Effort Estimation: Sinhal et al., International Journal of Advanced Research in Computer Science and Software Engineering 3(6), June - 2013, pp. 1120-1135 >@ %DUU\ %RHKP ʊ6RIWZDUH (QJLQHHULQJ (FRQRPLFV‫ۅ‬ Englewood Cliffs, NJ: Prentice-Hall, 1981. >@3XWQDP/+ʊ$*HQHUDO (PSLULFDO6ROXWLRQWRWKH0DFUR 6RIWZDUH 6L]LQJ DQG (VWLPDWLQJ 3UREOHP‫ ۅ‬,((( 7UDQVDFWLRQV RQ Software Engineering, pp. 345-361, 1978. >@% : %RHKP DQG 31  3DSDFFLR ´8QGHUVtanding and FRQWUROOLQJ VRIWZDUH FRVW´ ,((( 7UDQVDFWLRQ RQ VRIWZDUH engineering , vol. 14 no 10 Oct 1988. >@ 0 -RUJHQVHQ % %RHKP DQG 6 5LINLQ   ³6RIWZDUH Development Effort Estimation : Formal Models or Expert -XGJPHQW"´,(((6RIWZDUHYROQo. 2 pp. 14-19 >@ - :HQ 6 /L =@ 0 -RUJHQVHQ   ³$ review of studies on expert HVWLPDWLRQ RI VRIWZDUH GHYHORSPHQW HIIRUW´ -RXUQDO RI 6\VWHPV and Software, vol. 70, no. 1±2, pp. 37±60 [10] Trendowicz (2013), Software Cost Estimation, %HQFKPDUNLQJDQG5LVN$VVHVVPHQW6RIWZDUH'HFLVLRQ0DNHUV¶ Guide for Predictable Software Development. Springer Verlag. [11] Charu Singh, Amrendar Pratap, Abhishek Singhal (2014) ³(VWLPDWLRQ RI VRIWZDUH UHXVDELOLW\ IRU FRPSRQHQW EDVHG V\VWHPXVLQJ VRIW FRPSXWLQJ WHFKQLTXHV´ -1-4799-4236-7/14 IEEE. Pp. 788-794. [12] Moahmmed Algabri, Fahman Saeed, Hassan Mathkour, 1HMPHGGLQ 7DJRXJ   ³2SWLPL]DWLRQ RI VRIWFRPSXWLQJ &RVW (VWLPDWLRQXVLQJ *HQHWLF $OJRULWKP IRU1$6$VRIWZDUHSURMHFW´ 978-1-4799-7626-3/15 IEEE . >@ %DEXVND 5 ³)X]]\ 0RGHOLQJ )RU &RQWURO´ .OXZHU Academic Publishers, Dordrecht, 1999. >@,PDQ$WWDU]DGHKDQG6LHZ+RFN2Z³6RIWZDUH'HYHORSPHQW (IIRUW (VWLPDWLRQ %DVHG RQ D 1HZ )X]]\ /RJLF 0RGHO´ ,-&7( Vol. 1, No. 4,October 2009. >@ .LUWL 6HWK $UXQ 6KDUPD DQG $VKLVK 6HWK ³&RPSRQHQW Selection Efforts Estimation ± D )X]]\ /RJLF %DVHG $SSURDFK´ IJCSS, Vol 3, Issue 3, 2009, 210-215. >@ 0RVKRRG2PRODGH6DOLX ³$GDSWLYH )X]]\ /RJLF %DVHG )UDPHZRUN IRU 6RIWZDUH 'HYHORSPHQW (IIRUW 3UHGLFWLRQ´ .LQJ Fahd University of Petroleum & Minerals, April 2003. [17] Prasad Reddy P.V.G.D, Hari CH.V.M.K, Jagadeesh M., ³7DNDJL-Sugeno Fuzzy Logic for Software Cost Estimation Using )X]]\ 2SHUDWRU´ ,QWHUQDWLRQDO -RXUQDO RI 6RIWZDUH (QJLQHHULQJ Vol 4, No.1 January 2011. 108

Human Computer Interaction >@=KLZHL;X7DJKL0.KRVKJRIWDDU³,GHQWLILFDWLRQRI IXzzy PRGHOV RI VRIWZDUH FRVW HVWLPDWLRQ´ )X]]\ 6HWV DQG 6\VWHPV 2004, 141±163.

[20] Neetu Kushwahal 6XU\DNDQH   ³6RIWZDUH &RVW (VWLPDWLRQ XVLQJ WKH ,PSURYHG )X]]\ /RJLF )UDPHZRUN´ -14799-3064-7/14/$31. 00©20 14 IEEE

>@ 5DPD 6UHH    ³DQDO\WLFDO VWUXFWXUH RI D IX]]\ ORJLF FRQWUROOHU IRU VRIWZDUH GHYHORSPHQW HIIRUW HVWLPDWLRQ´ -14799-7678-2/15/$31.00 ©2015 IEEE

>@ ,PDQ $WWDU]DGHK 6LHZ +RFN 2Z   ³3URSRVLQJ D 1HZ +LJK 3HUIRUPDQFH 0RGHO IRU 6RIWZDUH &RVW (VWLPDWLRQ´ -07695-3925-6/09 $26.00 © 2009 IEEE DOI 10.1109/ICCEE.2009.97

109

Cognitive Knowledge Engineering

Chapter 23

Solving TSP Using Genetic Algorithm Approach Samreen Belim

S. N. Kakarwal

PES College of Engineering Aurangabad.

PES College of Engineering Aurangabad.

[email protected]

[email protected]

ABSTRACT

2. LITERATURE SURVEY

The main objective of the project is to solve the Travelling Sales 3HUVRQ¶V763 SUREOHPXVLQJWKHJHQHWLFDOJRULWKPDSSURDFK,Q this we have first of all studied the traditional approach for solving the TSP i.e. an Approximation algorithm. Then we have used an evaluating algorithm to solve TSP which is genetic algorithm. We have also studied the work that has been carried out on Genetic Algorithm (GA) and then used this approach to find optimal solutions. In the last, we have developed an application in .NET to solve TSP using graphical output.

The system that we have studied for the literature survey uses Approximation Algorithm. The best known approximation algorithm has been found in the seventies by Christofides[2]. The approach used in this algorithm is as follows:

General Terms Evolutionary Algorithms, Genetic Algorithm, Big Data.

Keywords Travelling Salesman Problem (TSP), Genetic Algorithm(GA), Genes and Chromosomes, Reproduction Operator and Mutation Operator.

1. INTRODUCTION 1.1 Problem Description We have a set of places to be visited and the cost of travel (or distance) between the possible pairs of places. The Travelling Salesman Problem is to find a travel plan, in which we will visit all the places and return to the starting point. The travelling plan should be made in such a way that, it will have minimum cost of travelling (or travel distance). The important point is the creation of populations, i.e. a set of all possible solution that can be generated for the given number of places. This is the main traits used by the Genetic Algorithms. Now we will try to implement a genetic algorithm with the purpose of solving the Travelling Salesman Problem that uses Genes and Chromosomes and also to implements the mutation and reproduction functionality.

1.2 Complexity 1.2.1 Number of Places The number of places to be visited is n and the total number of possible routes that will cover all the places will consist of a set of feasible solutions and can be given as (n-1)!/2 [1].

1.2.2 Number of Salesmen The application will create solutions for individual salesman. If two or more salesmen are supposed to visit the same places, they can follow the same solution generated.

1.2.3 Other Constraints Other constraints that can be used is the number of cities each salesman can visits, or the distance a salesman travels can be set to minimum or maximum, or any other constraints.

In this first of all a graph G is considered with an even degree on each vertex. It will then proceed on each connected part with more than one element. If no such part exists, then it is considered as an Eulerian graph. As it is an Eulerian graph so, for every vertex of this graph, there is an even number of edges. So now each vertex has pairs of such edges, one in and one out. Then, the out one is obviously an in of the next vertex. As it is an in, there is an out linked to it. And the process goes so on till we arrive back to the starting point. We continue the process till we visit each edge and no edge remains unvisited. Since, we have generated a connected graph; we can crumple all these tours into one unique order to create an Eulerian tour. Let us take any vertex as a0 in the Eulerian tour. For every ai of this path, it calls ai+1 the next vertex in the path that has not been visited yet, that means it satisfies the condition a i+1 DjIRUDOOM” i. Else it continues to find such a vertex in the Eulerian tour that has been not visited and stops after it has visited every vertex in WKHJUDSK1RZLWVKRZ¶VDSDWKRUDF\FOHsuch as aoíD1ííDn = a0, it is now termed as Hamiltonian tour. The graph is complete and so every pair (ai, ai+1). Furthermore, due to the triangle inequality, the cost of this edge is always less than the cost of the path between ai and ai+1 in the Eulerian graph. Therefore, it has Val (Eh ”9DO(e). Ee is the cost of Eulerian tour and Eh is the at most cost of Hamilton tour. Solving even moderate size of the TSP using such approximation algorithm requires vast computational time. The approximate approach never ensures an optimal solution but tries to bring close to optimal solution in a reasonable computational effort. It can be used when there are not much complicated data and the no. of nodes or cities are less. But as the data gets complicated, the time required to compute solution for such data also increases. This algorithm is very clear and elegant but in 40 years, it has not been able to improve it, even with something really sophisticated. On the other hand, it has analyzed the integrality gap of the relaxation in order to find a better approximation in another way. Even if all the specific cases we can think about do not exceed the 4/3 bound, the upper bound is still 3/2 for the general case [3].

3. GENETIC ALGORITHM 3.1 Related Work In the past years the TSP has been formulated as a search problem and had been answered by means of standard tree-search algorithms. Though the state space is very big (on the order of n! where n is the number of nodes in the graph) for such kind of 110

Human Computer Interaction problems. Genetic Algorithm, offers an well-organized method to search such big state spaces in parallel. The Genetic Algorithm falls in class of Evolutionary Algorithms. It is an optimization technique which searches a large space of candidate solutions in parallel. The algorithm starts to sample the search space with a set of individuals (called the population), and next it evaluates the quality of every individual and after that sampling a new set of individuals derived from merging good traits of such individuals from the earlier population. In our situation of TSP, one possible tour is represented by the chromosome and the cost of the tour is represented by the quality of individuals present in that chromosome. Low cost fragments of tour can be merged to form new individuals.

3.2 Introduction to Genetic Algorithms The Genetic algorithms (GAs) are actually based on the same theory of the survival of the fittest. When the species are produced by random transformation in the gene-structure of the chromosomes only the fittest and best species survive. The weaker one is generally lost by time. To solve any real life problem using GA, there are two conditions that should be satisfied: (a) A string that can represent a solution of the solution space, and (b)An objective function and hence a fitness function which measures the goodness of a solution that can be constructed / defined [4]. A simple GA works by randomly generating an initial population of strings, which is referred as gene pool and then applying (possibly three) operators to create new, and hopefully, better populations as successive generations. The first operator is reproduction operator, where strings are copied to the next generation with some probability based on their objective function value. In this not many changes appear in the new string generated. They are moreover like the old string. The second operator is crossover operator, where randomly selected pairs of strings are mated, and new strings are created. In this there is crossover between the old strings and the new strings generated differ from the old ones. The third operator, mutation, is the occasional random alteration of the value at a string position. The crossover operator jointly with reproduction operator is the largely significant process in the GA. Mutation diversifies the search space and protects from loss of genetic material that can be caused by reproduction and crossover.

3.3 Genetic Coding Each gene of a chromosome takes a label of node such that no node can appear twice in the same chromosome. So simply it means that the genes will directly represent the nodes or the cities to be visited. The chromosome will consist of the tour or the order in which the nodes could be visited. There are mainly two methods for representing tour of the TSP ± adjacency representation and path representation. We have considered the path representation method for a tour, in which it simply lists the label of nodes. For example, let {1, 2, 3, 4, 5} be the labels of QRGHVLQDQRGHLQVWDQFHWKHQDWRXU^ĺĺĺĺĺ` may be represented as (1, 3, 4, 2, 5).

3.4 Fitness Function The GA is generally used for maximization problem. For the maximization problem the fitness function is same as the objective function. But, for minimization problem, it is the reciprocal of REMHFWLYH IXQFWLRQ 7KH µILWQHVV IXQFWLRQ¶ FDQ EH GHILQHG DV follows

where f(x) is the objective function. Since, TSP is a minimization problem; we have considered this fitness function, where f(x) calculates cost (or value) of the tour represented by a chromosome.

3.5 Reproduction operator In reproduction/selection process, chromosomes are copied into next generation mating pool. The reproduction operator is associated with a probability with their fitness value. It helps to assign the next generation a higher portion of the highly fit chromosomes from the previous generation.

3.6 Sequential Constructive Crossover Operator (SCX) The search in the solution space is done by creating new chromosomes from the previous ones. Crossover is an important process in doing so. The sequential constructive crossover (SCX) operator constructs an offspring using better edges on the basis of WKHLUYDOXHVSUHVHQWLQWKHSDUHQW¶VVWUXFWXUH6RPHWLPHVLWLVDOVR able to generate better edges, which are not present in their SDUHQW¶VVWUXFWXUH7KH6&;GRHVQRWGHSHQGRQO\RQWKHSDUHQWs' structure; however it occasionally introduces new, although good, HGJHV WR WKH FKLOGUHQ ZKLFK ZHUH QRW SUHVHQW LQ WKH SDUHQW¶V population. A preliminary version of the operator is reported as local improvement technique.

3.7 Mutation operator Mutation operator selects a position in the chromosome randomly and changes the corresponding allele and so the newly formed chromosome is the modified one with the new set of information in it. Mutation on the chromosomes is done with the intention of supporting the reality that as the less fit members of the following generations are thrown away; some aspects of genetic material may be lost forever. To preserve this generic material and to maintain the diversity, some random changes in the chromosomes are performed. GA ensures to facilitate new-fangled parts of the search space are attained, while reproduction and crossover would not have been able to guarantee this. Mutation is an important trait in the GA that guarantees that no significant features are prematurely lost, hence preserving the mating pool diversity. In TSP, the classical mutation operator does not work. For this reason, we have considered the reciprocal exchange mutation that selects two nodes randomly and swaps them.

3.8 Genetic coding These are the parameters that govern the GA search process. (a) Population size: It determines how many chromosomes can be generated and so, how much genetic material is available for use during the search. If the genetic material is too little, the search has no chance to suitably cover the space. If there is too much, the GA wastes time evaluating chromosomes. (b) Crossover probability: It specifies the probability of crossover occurring between two chromosomes. (c) Mutation probability: It specifies the probability of doing bit-wise mutation. 111

Cognitive Knowledge Engineering (d) Termination criteria: It specifies when to terminate the genetic search.

4. DESIGN 4.1 Genes and chromosomes In the Travelling Salesman problem, the gene represents the Location that must be visited and the arrays of those Locations are represented by the chromosomes, which will successfully convey the travel plan i.e. the order in which the places must be visited. The size of those chromosomes is equivalent to the number of places (or locations) that must be visited. To obtain a valid chromosome, all the locations needs to be visited exactly once. So, the different travel plans will be represented by the chromosomes. The genes will represent the cities and chromosomes representing the different solution or paths possible. So, the best chromosome will be the one which will have the least distance travelling all the nodes, and would be the best possible solution. My basic idea is not to always find the best chromosome, but to select the optimal one from the population being generated.

4.2 Permutations Permutation function is used to produce the mutation of chromosomes which can be used later on. We have considered three types of permutations i.e. the swaps, the moves and the reverse ranges. We have used these in our application with the purpose to generate better solutions in less steps so that it requires less time. Certainly those changes are made randomly and at times it may not generate satisfying solutions, however it is always likely to generate a improved result with mutations at the correct position with no need of the intermediate results.

Our actual idea is to validate a gene as soon as it does the crossover. So, we have created a hashset with all available locations, and then we have used a foreach for all the possible Locations in the chromosome, later than we have either removed these locations from the hashset or, tried to keep a record that such a gene must be replaced by an available one. First of all we have completed the loop and then, if there are genes that can be replaced, we have simply replaced them by the ones that are still available. Maybe it will not create the best possible solution, yet we have tried creating "travelling paths" that does not visit the same locations twice and forget to visit others.

5. IMPLEMENTATION 5.1 Genes and chromosomes Once we have constructed a hashset, a preliminary population of chromosomes (the travel plans) must be created. The locations could be visited randomly, but we have used an algorithm instead for doing so. After initial population is created, the code enters a loop in which: 1. The best chromosomes are selected (the travel plans); 2. Makes an effort to offer the best solution possible at that moment; 3. Also it is capable of reproducing them and carries on the loop with the next generation. The factors that are certain to this problem are: 1. How it comes to know which the best chromosomes are? 2. What are the criteria to terminate? 3. During the reproduction, should crossovers be used, or mutations or what exactly to evolve?

At times chromosomes are generated using all three operations and thus the best chromosome is selected as the solution.

My attempt to solve these problems is: 1. It just computes the entire distance traveled taking into account the order in which the locations are visited. The start and the end location is the same, so it is located on the top-left. To prevent wastage of time this is not encoded in the chromosomes, as the chromosomes might start at the wrong place, however it is added in the total distance travelled. 2. The algorithm in no way is able to stop by itself. The user of the application is responsible for deciding when to stop considering the best solution tat was found. 3. Doing the reproduction phase, we have considered only half of best chromosomes leaving the remaining half that will not reproduce. It can use either crossovers or random mutations that will in the end develop good solutions however it might spend plenty of time generating wrong solutions.

4.3 Crossovers

5.2 Explanation of Key functions

Studies in GA suggest us to facilitate a higher use of crossovers and merely some mutations. Although using crossovers in our TSP problem, we might visit the same location twice and consequently others might remain not visited. So, should we use crossovers in the Travelling Salesman Problem?

5.2.1 Main Window

6RPHWLPHV ³VZDSV´ FDQ EH XVHG LQ WKH SUHVHQW VROXWLRQ LH QRW just randomly visiting the nodes, but just swapping the order of YLVLWLQJWKHQRGHVWRPLQLPL]HWKHGLVWDQFH7KXV ZHGRQ¶WKDYH to change the entire path, but just the position of two nodes gets LQWHUFKDQJHG ,Q ³PRYH´ RSHUDWLRQ WKH LWHPV DUH VKLIWHG E\ RQH position i.e. the first item become the last one, the second item becomes the first one and so on. This was not possible with single swap and so we decided to add such "move" operation. Here, DOPRVW WKH FRPSOHWH SDWK JHWV FKDQJHG VRPHWLPHV ,Q ³UHYHUVH UDQJH´ZHMXVWYLVLWWKHQRGHVLQWKHUHYHUVHRUGHUKRSLQJWRILQG a better solution.

Explaining it, yes, we have used crossovers, and made an attempt to prevent these repeated locations. The repeated locations problem should be solved correctly while using random mutations. Suppose we have to visit ABCDE, but the genes can visit around $$$$$6RIRUQRZLWKDVYLVLWHGORFDWLRQ³$´ILYHWLPHV For this we have created a list of available location. So, when a particular location is visited, it is removed from the list of available locations. Now possibly the genes will visit the locations that were named as ABCDE.

This function initializes the software and provides a graph area to draw the output graph with the result. This is the most important form in the application. It has all the functions provided as checkbox and buttons. So, the user needs not to navigate to other forms. He can perform all the functions on the main form itself. This form has the coding in xaml i.e eX-tensible Application Markup Language and cs i.e. C-Sharp. This form has both the extensions i.e. MainWindow.xaml.cs.

5.2.2 Location This function is used to mark the locations on the graph, so that destinations can be formed and then distances between the pixels can be found. It is implemented in C-Sharp..

112

Human Computer Interaction

5.2.3 Random Provider This function generates random values for x and y coordinate, so that different location can be formed i.e. the genes are generated here. The number of locations is also decided and can be increased if required. The output of this form is provided to Location.cs form, to generate the random destinations.

5.2.4 Travelling Salesman Algorithm This is the most important form in the application, as it implements the actual algorithm. All the functions needed to generate the output are implemented in this form such as move, reverse, mutation and crossovers. Also the chromosomes are generated here. This form is implemented in C-Sharp,. This form fetches the destination from Location.cs form, generates the output and provides it to the Main.xaml.cs form to represent it in the graphical output.

Snapshot 3 : Distance = 1793

6. RESULT Below are the snapshots of application created in Visual Studio 2010 using C# language. There are 6 locations taken and the different tours are created for this. The starting and ending point is the same, which is in the top-left corner of the snap. The distance between the locations is taken in the pixels. The total distance for the tour is calculated and written in top-left corner of the snap. Here, we have taken 9 different tours.

Snapshot 4 : Distance = 1806

Snapshot 1 : Distance = 1822

Snapshot 5 : Distance = 1816

Snapshot 2 : Distance = 1854

Snapshot 6 : Distance = 1819

113

Cognitive Knowledge Engineering

Snapshot 3

1793

Snapshot 4

1806

Snapshot 5

1816

Snapshot 6

1819

Snapshot 7

1875

Snapshot 8

1768

Snapshot 9

1849

On analyzing the result from the above table, 1768 pixels i.e. Snapshot 8 is the shortest possible tour. Snapshot 7 : Distance = 1875

7. CONCLUSION The application that we have developed is able to find results by means of mutations only, thus the crossovers can be ignored. Surely we have developed a basic application, not including most SURIHVVLRQDO SRLQWV EXW DV LW¶V D VLPSOHU RQH LW LV HDV\ WR understand and also it is able to find the optimal solution for most of the situations.

Snapshot 8 : Distance = 1768

Besides this, the algorithm rarely interacts with the user. This means that, the user only gives the number of destination and the algorithm finds out the shortest path. The user is not able to set the distances between the two destinations. Also it is the user who would have to decide when to stop and also to decide the optimal solution from the population being generated.

8. ACKNOWLEDGMENTS We are thankful to Prof. V. A. Losarwar for her affectionate encouragement, suggestions and valuable feedback on the paper.

9. REFERENCES [1] Traveling Salesman Problem: An Overview of Applications, Formulations and Solution Approaches ;Rajesh Matai1, Surya Prakash Singh and Murari Lal Mittal, Management Group, BITS-Pilani; Department of Management Studies, Indian Institute of Technology Delhi, New Delhi; Department of Mechanical Engineering, Malviya National Institute of Technology Jaipur; India.. . Snapshot 9 : Distance = 1849 Table 1. Distance in pixel Snapshot No.

Top

Snapshot 1

1822

Snapshot 2

1854

[2] Nicos Christofides. Worst-case analysis of a new heuristic for the travelling salesman problem. Technical Report 388, Graduate School of Industrial Administration, CMU, 1976.. [3] David B. Shmoys and David P.Williamson. Analyzing the held-karp tsp bound: a monotonicity property with application. Information Processing Letters, 35(6):281±285, 1990. [4] CS 6601 - A Artificial Intelligence Project 1 Report A Genetic Algorithm for Symmetric TSP.

114

Human Computer Interaction

Chapter 24

Behavior of Electromagnetic Waves at Antenna Boundary in Wireless Communication System: A tutorial Khan Sohelrana

Sayyad Ajij D.

Dr. B.A.M.University Aurangabad, M.S. (India)

Dept. of ETC MIT, Aurangabad, M.S. (India)

ABSTRACT This paper presents a study of electric and magnetic fields together at the antenna boundary using 3-D atmosphere which is easy for understanding. Static fieldV GRQ¶W FKDQJH ZLWK UHVSHFW WRWLPH ,Q few situations it is possible to treat the electric field and magnetic fields independently, without worrying about interactions between the fields. But the second category includes situation in which the fields varies with respect to time, and in such case it is not possible to treat the fields independently. This paper is a study of a changing electric field as a source of magnetic field. Whenever either field changes with time, a field of the other kind is induced in adjacent regions of space. Electric field intensity and Magnetic field intensity moves together in space and are of time-varying nature. Electric and magnetic fields bounces back if strike on a conductor. This tutorial presents wave propagation through space from one region to another using 3-D set-up. Further it uses SWIPT with antenna in wireless communication system as an example for explanation.

General Terms Antenna EMF radiation

Keywords Standing wave, EMF, SAR, Antenna, Boundary Conditions

1. INTRODUCTION

necessity of exploring these terms with respect to antennas used in wireless communication. In this paper we first explain the basic behavior of EMF wave in detail, using 3-D diagrams. Later we discuss EMF behavior and its relativity with simultaneous wireless information and power transmission (SWIPT). Which is combination of wireless power transfer (WPT) and wireless information transfer (WIT) related terms. In [1] many recent advances in WPT are reported in detail. The relation between WPT and EM fields are also studied in various forms using antenna theory for understanding the EMF effects on humans [2][3][4][5]. Specifically in [4] SAR effects on human eyes due to the field radiated by user antenna in wireless LAN are given EMF aware cell selection in wireless communication is given in [6]. Whenever a text message or a call from cell-SKRQH LV PDGH WKHQ WKHUH LV WUDQVIHU RI µSURSDJDWHG information from one location to another. For this information transmission there are three basic types of propagation namely; 1) Ground wave propagation. 2) Space wave propagation. And 3) sky wave propagation. Various methods are shown in fig. 1. In this paper we are using images created in 3-D atmosphere. The usage of 3D diagrams is very useful to understand and convey the physical interpretation of EMF terms and behavior [7][8]. EMF waves are transverse waves. Transverse wave is a moving wave which consists of oscillations perpendicular to the direction of flow. If a transverse wave moves in positive x-direction, its oscillations move in up and down directions, which look like ocean waves. A transverse wave is a moving wave which consists of oscillations perpendicular to the direction of flow. Voice is longitudinal in nature, and when the conversation is transmitted from the cell phone then it is of transverse nature.

In any wireless communication, antenna is an important parameter. It is important to study the behavior of electromagnetic fields (EMF) around an antenna as the radiation transmitted/received by antenna is EMF wave. Moreover, as data usage is increasing in wireless communication, the radiation effects on health are becoming important topic to be investigated. The region which is close to antenna is termed as non-radiatingnear field. Outside this region there is presence of radiating far field where intensity is inversely proportional to square distance. Wireless communication research is enhancing in near [11] as well as far field radiation applications.

Light wave is same as of electromagnetic wave as shown in Fig. 1. The magnitudes of the field vectors E and B are in phase and are related by E=cB, with (1)

Recently, Energy harvesting (EH) in wireless communication attracted great concentration of researchers. EH using various methods are being studied and proposed. Power allocation in battery constraint systems is the main issue present in this area. Due to this requirement, we are working on the power allocation strategies in energy harvesting system using EMF behavior in wireless communication. We are trying to investigate, whether it is possible to create the power allocation method, which can be used to reduce the radiation effect by considering various parameters in EH system with poynting vector as power flow parameter using study of EMF behavior. As radiation and SAR are related to power density [12-13] which is direct function of electric field intensity (EFI) and magnetic field intensity (MFI) we find the

And c =2.9979246 x 108 m.s-1 is the speed of light in the vacuum. are the permeability and permittivity (i.e., the dielectric constant) of the vacuum. As distance increases the wave becomes flat and ends sometime later. These waves can also be used as energy harvesting element in ractennas [9]. James Clerk Maxwell first described electromagnetic waves. Calculations led Maxwell to conclude that light itself is an EM wave. Speed of light and EM wave is same. This visible part of the electromagnetic spectrum consists of the colors that we can see in a rainbow. Red, orange, yellow, green blue, indigo and violet are the colors. Frequency is nothing but the number of waves that pass a fixed place in a given time. Light usually has a spectrum of frequencies which sum together to form the resultant wave. Different frequencies undergo

2. LIGHT AS A WAVE

115

Cognitive Knowledge Engineering different angles of refraction Maxwell stated that a time-varying electric field generates a time-varying magnetic field and viceversa. These changing fields together form a propagating electromagnetic wave. The maximum disturbance, or height of a wave, is called its amplitude. The distance from crest to crest or trough to trough is gives the wavelength .

3. WAVE EQUATION In nature waves are present everywhere and invisible to human eyes. Appropriate gadget is required to access them. These waves travel across free space, dielectrics and almost everything. The wave emerges from the antenna of sender and terminates in the antenna of receiver. Waves behave differently in different mediums according to behavior of that medium. In initial conditions this paper considers that waves are lossless. And they may become weak after a certain time and distance. Basic relations related to electromagnetics are: Fig. 1. Electromagnetic Wave as light wave Up and down movement of wave can be resembled with ocean wave (Electric field (E) wave in this case) and side by side movement with snake (magnetic field (H) wave in this case). Lorentz equation gives relation of force and electro-magnetic terms as shown in fig 2. Electric field intensity is defined as;

Where D & B are densities of E and H respectively. In Cartesian coordinates Del operator is given as;

Physical wave must be converted in mathematical equation. TableEHORZJLYHV0D[ZHOO¶V(TXDWLRQ>@ Field equation in (2) is true for static condition. But when we consider the movement then we have to consider velocities associated with that movement.

Details

TABLE 1 0$;:(//¶6 EQUATION In the form of operator

Integral form

*DXVV¶VODZIRU electric term *DXVV¶VODZIRU magnetic term 3.Faradays law of induction 4.Amperes law

Fig. 2. Pictorial representation of relation between F & Q Transverse waves exhibits a phenomenon called polarization. A wave moving vertically up and down can be termed as linearly polarized wave. A wave circulating in a circle is a circularly polarized wave. Similarly, there may be elliptically polarized wave. Electric field intensity on charge Q is greater than the effect of magnetic field intensity H. E and H vectors are perpendicular to the direction of propagation of the radiation. Speed of wave is given by;

Wave equations can be given for: Conductors, Free space, lossless dielectrics, lossy dielectrics, good conductors etc. Various equations can be derived by using simple considerations like: For Conductors: For Free space: For lossless dielectrics:

Distance between crest to crest is nothing but wavelength. And time is the period of wave.

For lossy dielectrics : 116

Human Computer Interaction In this paper we are exploring boundary conditions across antenna. Antenna surface is conducting medium and surrounding region is air. So, in next sub-section we discuss wave equations for conductors and dialectics.

Rearranging terms;

3.1 Wave Equations for Good Conductors In this section wave equation for free space is given where given as conductivity,

Keeping values of J,

permittivity, and

is

as permeability.

in Table-1 we get ;

Fig.3. SWIPT- Wireless Energy Harvesting

For a source-free region we have; and

From vector calculus, vector identity, Equation 3 is termed as wave equation for E. In same way equation for H is given by. Where

is any vector. For vector EFI we can write; This will lead to equation 4 as shown below;

From equation (A) Equation 3 & 4 are wave equations for conductors. 117

Cognitive Knowledge Engineering

3.2 Wave Equations for Free Space In this tutorial we imagine wave propagation in free space which starts from transmitting antenna. Conductivity for free space is , then obviously zero. So, wave equations reduces to;

Fig. 4. Electric Field Vector )URP0D[ZHOO¶VHTXDWLRQJLYHQLQWDEOH-1.

These are wave equations of free space. In similar way one can change conductivity, permittivity and permeability values to get wave equations of different mediums.

4. RELATION BETWEEN E & H Fig. 3 shows a SWIPT system where RF energy is used as WIT and EH element. It set ID=0 for information decoding and EH=1 for energy harvesting. However, simultaneous transfer of WIT & WPT is difficult due to practical circuit constraints. In future it may be possible that reflected power from antenna surface can be captured for additional WEH. In fig.3 it is shown that when a wave strikes on the surface of a medium (antenna in this case) then a standing wave generates which reflect back and changes amplitude of incident wave. It may also be possible that this reflecting wave can be utilized as the EH source signal before reflecting. This can serve dual benefit of storing energy as well as betterment in VSWR value of the signal. It is possible to study behavior of Poynting Vector on the boundary of a receiving antenna. Propagation of wave is considered in x-direction. Region . This region is free space. Region Z Z < 0 is medium-1 with > 0 is medium-2, which is receiving antenna medium as shown in direction, H is vibrating in direction. fig.3. E is vibrating in

Vibrations of H are as shown in fig.5 this wave is making half circle pattern as shown in figure.

E and H components are moving in direction. In this tutorial we consider the relationship between E and H components for the forward traveling wave. Here a uniform plane wave is considered. Uniform plane wave is basically a wave that possesses constant propagation without any change in amplitude as shown in fig.1. It may be noted that the field components are functions of both space (x, y, z) and time (t). For example, in a Cartesian coordinate represented as and . /HW¶V system, consider a plane wave that is moving in X-direction. In this case, travelling wave in x-direction is given by: Fig. 5.Magnetic Field intensity part of total EM Wave /HW¶VFRQVLGHU(DQG+FRmponent separately. Fig. 4 explores this equation.

in similar way :

118

Human Computer Interaction

This equation 6 is explained in fig.6.

Fig. 6. Pictorial Explanation of Equation-6 E and H always travel together in same direction with perpendicular relation. This perpendicular movement is termed as Transverse and EMF is termed as TEM. Equations derived from earlier expressions are:

Fig.7. Physical significance of equation 7 This equation is shown in fig.7 is for two instances of time. This wave can be given by:

From earlier discussion;

The above equation has a solution of the form :

Keeping value of

we get:

In the absence of any reflection, the second form of the above eqn is zero and the solution can be written as

Since, it is function of one direction only we can take partial term as;

119

Cognitive Knowledge Engineering

The constant of integration represent that this field is independent of x may also exist. However, this field will not be a part of the wave motion.

It gives travelling relation of E and H as

Above term is called as characteristic or intrinsic impendence of the free space

Fig.9. Representation D component across border It can be observed from fig. 8 & 9 that charges are distributed on WKH VXUIDFH RI D FRQGXFWRU 7KLV UHSUHVHQWV WKDW µ(¶ DQG µ4¶ LV DOZD\V]HURLQVLGHDFRQGXFWRU¶DQGDSSHDUVRQVXUIDFHRQO\7KLV effect cancels line c-d component. i.e.

5. BOUNDARY CONDITIONS FOR ANTENNA

At border line E have normal and tangent components. a) Tangent boundary conditions:

Antenna transmits EMF waves in surrounding region. The boundary conditions between antenna and air are discussed in this section. For solving these problems we require the knowledge of the relations of field quantities at an interface between two media that is nothing but border or boundary. We have;

For total area:

b) Normal tangent component: For normal component Gaussian surface is considered as shown in ILJ$SSO\LQJJXDVV¶VODZIRUJLYHQILJXUHZHJHW

Fig. 8 Boundary between conductor (antenna) & dielectric Consider a closed path ± a b c d as shown in fig.8. In figure it can be seen that border line is separating two different types of materials. Here, all normal and tangent components of E are considered. E for closed path a- b- c- d is due to combination of four lines. ab, bc, cd, and da.

Cylinder is combination of top, bottom and sides surface as shown in fig. 9 ;

/HW¶VFKHFNDOOVLGHVRQHE\RQH i) Top:

120

Human Computer Interaction

Fig. 10. A 3-D Exploration of Change in the signal due to standing wave effect ii)

Bottom:

Since iii)

Sides:

Since, D exists only on top surface. Another important parameter related to antennas is VSWR. Ideal YDOXHRI96:5LVµ¶DVLWLVWKHUDWLRRIPDx and min values of voltage as shown in fig. 10. Practically there will be change in Vmax and Vmin, if the signal which reflects back from the receiver antenna mixes with the incident wave as shown in fig. 10. Fig. 3 explains one of the optimistic solutions for this problem in ideal conditions.

6. FURURE WORK It is be possible to investigate power allocation strategy based on EMF behavior in EH wireless environment where nodes/mobiles are situated in dense atmosphere with problems associated with EM Radiation. We strongly believe that power allocation based on the study of electromagnetic field behavior will certainly help in utilizing power efficiently to minimize the power consumption by node to enhance battery life and reduction in radiation based problems.

7. CONCLUSION This paper illustrates equations and expression for characteristic impedance of a wave using 3-D environment. The EMF behavior across the antenna border is discussed. Physical characteristics of waves are explained in terms of numerous diagrams. The wave behavior is important in understanding the cellular/mobile communication as the basic element of wireless communication. A wave behavior is easily explained for non-technical readers who are curious to understand the various terminologies related to cellular mobile communication. Importance of EMF with respect to EH wireless communication system is discussed. SWIPT & possibilities of power allocation strategies based on EMF behavior are briefed.

8. REFERENCES [1] S. Y. R. Hui, Wenxing Zhong, and C. . /HH ³A Critical Review of Recent Progress in Mid-Range Wireless Power

Transfer´ LQ IEEE Transactions On Power Electronics, Vol. 29, No. 9, September 2014, pp. 4500-4511. [2] Seong-Min Kim, Jung-Ick Moon, In-Kui Cho, Jae-Hun Yoon, Woo-Jin Byun, and Hyun-&KXO &KRL ³$GYDQFHG 3RZHU Control Scheme in Wireless Power Transmission for Human 3URWHFWLRQ )URP (0 )LHOG´ LQ IEEE Transactions On Microwave Theory And Techniques, Vol.63,No. 3, March 2015, pp. 847-855. [3] Chung-Huan Li, Mark Douglas, Erdem Ofli, Nicolas &KDYDQQHV 4XLULQR%DO]DQR DQG 1LHOV .XVWHU ³0HFKDQLVPV of RF Electromagnetic Field Absorption in Human Hands and )LQJHUV´ LQ IEEE Transactions On Microwave Theory And Techniques, Vol. 60, No. 7, July 2012, pp.2267-2276. [4] Paolo Bernardi, Marta Cavagnaro, Stefano Pisa, and Emanuele 3LX]]L ³6$5 'LVWULEXWLRQ DQG 7HPSHUDWXUH ,QFUHDVH LQ DQ Anatomical Model of the Human Eye Exposed to the Field 5DGLDWHG E\ WKH 8VHU $QWHQQD LQ D :LUHOHVV /$1´ LQ IEEE Transactions On Microwave Theory And Techniques, Vol. 46, No. 12, December 1998, pp.2074-2082. >@ 5RQROG : 3 .LQJ ³(OHFWULF &XUUHQW DQG (OHFWULF )LHOG Induced in the Human Body When Exposed to an Incident (OHFWULF )LHOG 1HDU WKH 5HVRQDQW )UHTXHQF\´ LQ IEEE Transactions On Microwave Theory And Techniques, Vol. 48, No. 9, September 2000, pp.1537-1543. [6] Antonio De Domenico, Luis-Francisco Diez, Ramón Aguero, 'LPLWUL .WpQDV DQG 9DOHQWLQ 6DYLQ ³(0)-Aware Cell 6HOHFWLRQ LQ +HWHURJHQHRXV &HOOXODU 1HWZRUNV´ LQ IEEE Communications Letters, Vol. 19, No. 2, February 2015, pp.271-274. [7] Khan SohelRana, Sayyad Ajij D and Mohammed Laeeq 6KDLNK ³Survey, Study & Review of Coordinate System Notations with the Use of 3D Environment for Better Understanding of EM Fields Behavior´LQ13th IFIP & IEEE International conference on Wireless and Optical Communication Networks, Hyderabad, India, 22-25-July-2016. [8] C. Bachiller, H. Esteban, S. Cogollos, A. San Blaq and V. E. %RULD ³7HDFKLQJ RI :DYH 3URSDJDWLRQ 3KHQRPHQD 8VLQJ 0$ 7/$%*8O¶VDWWKH8QLYHUVLGDG3GLWHFQLFDRI 9DOHQFLD´ in IEEE Antennas ond Propagation Magazine, Vol. 45. NO. I, February 2003, pp.140-143. [9] Hannes Reinisch, Stefan Gruber, Hartwig Unterassinger,, 0DUWLQ :LHVVIOHFNHU HWDO ³$Q (OHFWUR-Magnetic Energy Harvesting System With 190 nW Idle Mode Power &RQVXPSWLRQ IRU D %$: %DVHG :LUHOHVV 6HQVRU 1RGH´ LQ IEEE Journal Of Solid-State Circuits, Vol. 46, No. 7, July 2011, pp.1728-1741. >@ -RKQ ' .UDXV 5RQDOG - 0DUKHIND $KPDG 6 .KDQ ³ $QWHQQD EDVLFV´ DQG ³ 5DGLDWLRQ´ LQ Antennas and wave propagation, Fourth Edition, Special Indian edition, Tata McGraw Hill Education Private Limited, New Delhi, 2012, Pg. 8-45 and 69-82. >@.KDQ6RKHOUDQD6D\\DG$MLM'³Design and Simulation of NFC Printed Antenna in Near Field Wireless Communication IRU 6'&´, IEEE International Conference on Distributed Computing, VLSI, Electrical Circuits and Robotics at, NITK Surathkal, Mangalore Karnataka INDIA, 13-14 August 2016. [12] Focus group report, ITU-T,FG-66&³(0)FRQVLGHUDWLRQVLQ VPDUWVXVWDLQDEOHFLWLHV´ >@,&1,536WDWHPHQWRQWKHµ*XLGHOLQHVIRUOLPLWLQJH[SRVXUHWR time-varying electric, magnetic, and electromagnetic fields (up WR  *+] ¶ ,&1,53 +HDOWK 3K\VLFV  -258, September 2009.

121

Cognitive Knowledge Engineering

Chapter 25

A Survey on Vision-based Hand Gesture Recognition Sunil G. Deshmukh

S. M. Jagade

Department of E &Tc Marathwada Institute of Technology, Aurangabad Maharashtra, India

College of Engineering Osmanabad Mahrashtra, India

ABSTRACT Gesture recognition acts very important role in human and computer interactions. The main objective of gesture recognition is to develop the system to recognize specific hand gestures and use it to convey the information. In this paper a survey on various latest tools for gesture recognition is given with the various input devices and algorithms is used. A review on dynamic hand gesture recognition process, vision based dynamic hand gesture recognition system, segmentation and feature extraction methods, hidden Morkov model and challenges of dynamic hand gestures are highlighted.

General Terms Hand gesture recognition Systems, Human computer Interface, Hand gestures recognition tools.

Keywords Gesture Recognition, Segmentation and feature extraction methods, Dynamic hand gesture recognition process.

1. INTRODUCTION Hand gesture recognition is used to create user interfaces that are natural to use and easy to learn [50]. Sensors used for hand gesture recognition include wearable sensors such as data gloves and external sensors such as video cameras. Data gloves can provide accurate measurements of hand pose and movement, but they require extensive calibration, restrict natural hand movement, and are often very expensive. Video-based gesture recognition addresses these issues, but presents a new problem: locating the hands and segmenting them from the background in an image sequence is a non-trivial task, in particular when there are occlusions, lighting changes, rapid motion, or other skin-colored objects in a scene (see Mitra [33] and Erol [34] for reviews of video-based methods). This paper surveys numerous papers that describe gesture types and classification to answer two questions

x

x

What methods are being used to achieve hand localization and gesture recognition with cameras? Note that hand localization is in this context a computer vision problem and gesture recognition is a machine learning problem. What applications and environments are researchers testing their methods in? Do they test them in situations where their methods have supposed advantages over other video-based methods? Are the limitations of vision -based systems tested?

This survey is organized according to our novel conceptual model of the major components of a hand gesture recognition system, shown in Figure 1. Vision-based hand Gesture recognition begins with image acquisition, which is dependent on the sensor being used (Sec. 2.1). Then hand localization is performed on the acquired image sequence using tracking and segmentation methods (Sec. 3.1). Finally, the segmented hand images and/or their tracked trajectories are classified as a particular gesture or pose. (Sec. 3.2). Note that this review focuses on the components, involved in gesture classification, which is the process that recognizes a set of poses and gestures from a predetermined gesture set .Sec 4 discusses the Limitation of current applications developed for Hand gesture recognition. Sec. 5 discusses the applications reported to date, and Sec.6 details the conclusions

2. TOOLS FOR GESTURE RECOGNITION Gestures can be anything from bodily motion ranging from any movement or pause state but commonly face and hand movements are at core of gesture recognition. Gesture recognition is study of tracking person's movements with the help of various tools and then determine which abstract gestures they may be performing . Some common Input devices and common approaches are discussed below although some of these methodology's follow similar implementations.

2.1 INPUT DEVICES 2.1.1 Cyber Glove. These type of devices are used for very sophisticated systems which can provide accurate position and rotation of the hands using optical or magnetic markers and tracking of hand is obtain by inertial tracking devices in 3 dimensional space .Some gloves can monitor individual finger phalanges( Distal ,intermediate and proximal ) movement achieving 5 degree rotation detection accuracy in finger. The Data Glove[24] was first commercially available hand-tracking device available in 1989 by Young L. Harvill . But it was popularized by power glove by Nintendo.

2.1.2

Depth-aware cameras.

Using special cameras which use grid of light or laser pulse to measure the depth of room , user can generate a depth map of room at short range which then is used as to approximate a 3d representation of what is being seen. These can be useful in detection of hand gestures but it have short range capabilities [1].

2.1.3 Stereo cameras. Using two cameras to simulate human binocular vision .Once the two images captured at same moment, we can approximated 3d representation of image . To get the cameras' relations, lexian122

Human Computer Interaction stripe or infrared emitters can be used as positioning reference for Still Image .[2] This method is useful in Skeletal based gesture detection as it gives us direct motion measurement(6D-Vision) when used in combination .

2.1.4 Gesture-based controllers. Unlike Cyber Glove these controllers are hand held or wearable devices on body so that when movements are performed, their motion can be captured by monitoring software which will convert these movements to desirable gestures . these gesture controllers ranges from simple mouse movements to the Wii Remote or to the My armband or the Force Wizard wristband, which can detect changes in acceleration to represent gestures[3][4]. while Some devices use the Hillcrest Labs' Free space technology like LG Electronics Magic Wand, the Loop and the Scoop use to provide input to devices , These devices use MEMS accelerometers, gyroscopes and other sensors for cursor movement. The smart software can automatically calibrate human tremor and inadvertent movement for imprecise input [3][4] Audio Cubes are another example. The optical sensors of these cubes can be used to sense hands and fingers , and can be used to process data. it is widely used sound synthesis,[5] but can be applied to other fields.

2.1.5 Single camera. wide availability of standard RGB sensor camera turning to be most lucrative area of research in gesture reorganization using single camera and Software-based gesture recognition technology . Use of single camera also fit in when we have to overcome the convenient controlled environment used for gesture recognition. Though there are wide range of challenges such as capturing device quality is not standardize ,light illumines , differentiating background but this challenges can easily overcome by applying better algorithms some of these algorithms are discussed in section 3 .

2.1.6 Radar like conventional radar which uses segment of electromagnetic spectrum to beam the signals on objects then analyzing the reflected signal for object analysis ,but radar technology we know today is not portable it means that it uses very big equipments , also uses very dangerous electromagnetic waves which emit heat radiation as well as electromagnetic radiation . soli [6]Sensor technology works manage to use safe electromagnetic waves in a broad beam, which manage to eliminate all the above drawbacks and benefits of radar technology . It uses the returned signals to gain data about tKHREMHFW¶VVKDSH which can be used in gesture reorganization.

2.2.1 3D Model-Based Algorithms The 3D model approach which is actively used in gaming industry and animation industry for rendering virtual 3d space which can interact with user . 3D model used are normally volumetric models . The actual 3D model are made from 3D surfaces, like NURBS or polygon meshes.[8] these sub-3D surfaces provide flexibility and less computational complexity . Heavy computational cost using this method prevent the system to be developed for real time applications .But we can use simple approach to map simple primitive objects to the subjects most important body parts and analyses the way these interact with each other. Same way we can divide the whole body in to smaller parts and study them . As 3D model is easy to mimic the real world we can structure whole body as hierarchical object and apply constraints .

2.2.2 Skeletal-based Algorithms Skeletal-based algorithms use joint angle parameters along with segment lengths . The parts of the body are mapped on virtual skeleton . The virtual skeleton uses topography and acclimatization of body parts to relate with each other [9].Advantages of using skeletal models:

x x x

Abstract parameters gives faster results . Stored features are easy to match with . We can focus on certain parts of body for reorganization purposes .

2.2.3 Appearance-based models Instead of using spatial representation of the body, the appearance based model use the parameters from the images or videos by means of using a template to match database . Some of these methods are use templates of the human parts of the body. Templates are constructed using outline of body part then approximating them using linear interpolation to match with wide verity of inputs [10]. A second approach in gesture detecting using appearance-based models uses dynamic sequences as gesture templates to convey the complex messages .we can use images , or certain features derived from these images to derive the template for reorganization .

2.2 APPROACH The above input Devices provide us different type data to be interpreted, which can be processed in different way. Each system have its own merits. However, most of systems represent the data on different coordinate system the approaches for processing this data produces the results with high accuracy, In order to interpret movements of the body, one has to use pre-defined properties and the movements which can be used to express the common output . For example, in sign language each gesture represents a word or phrase. The taxonomy that seems very appropriate for Human-Computer Interaction has been proposed by Quek [7]. He presents several interactive gesture systems in order to capture the whole space of the gestures .

Fig 1: Dynamic Hand Gesture Recognition Process

3. VISION-BASED DYNAMIC HAND GESTURE RECOGNITION In vision based hand gesture recognition system, the video camera is used for capturing the motion of hand. This input video is 123

Cognitive Knowledge Engineering further processed to extract features by processing individual frame [10]. Pre-processing may be required to highlight the necessary components [11]. For example tracking the continues movement of hand in stable background is good approach for removing the unnecessary data from frame [10].Fig. 1 specify the general Dynamic Hand Gesture Recognition Process

3.1 SEGMENTATION AND FEATURE EXTRACTION METHODS The problem of locating the hands in an image is essential to gesture recognition and is split into two sub-problems: hand segmentation, and hand tracking. Hand segmentation is the problem of determining which pixels in an image belong to a hand, and hand tracking is the problem of determining the trajectory of a hand in a sequence of images.

3.1.4 OTHER METHODS Less general approaches to hand segmentation can take advantage of certain conditions in an application. Jojic [39] exploits a special case where the appearance of the background is known, and so uses static background subtraction for segmentation. Park [40], Raheja [41] and Yang [14] ask the user to wave their hand at the beginning of an interaction and use motion images (the accumulated difference betwHHQVXFFHVVLYHIUDPHV WRGHWHUPLQHWKHKDQG¶VORFDWLRQDVDQ area of high motion then use depth thresholding at that point for segmentation. And Wang [28] and Feris [42] use shadow analysis to find depth discontinuities and thus create a silhouette of the hand; however, this method only works well in very controlled lighting.

3.2 RECOGNITION METHODS 3.1.1 Motion Based Methods The tracking algorithms used were the Kalman filter (used by Park [25] and Trigueiros [26]), CAMSHIFT (used by Yoo [27] and Yang [14]), and mean shift (used by Chen [29] and Keskin [30]). The Kalman filter is a recursive least squares estimate of the state of a dynamic system. For gesture recognition, the state being estimated is the position and orientation of the hand in subsequent frames. Mean shift is an iterative mode-finding algorithm that uses gradient descent to estimate the direction and velocity an arHD¶V movement (a hand in this situation). Mean shift works well for deformable objects, making it well suited for hand tracking. CAMSHIFT (continuous adaptive mean shift) is an extension of mean shift that dynamically adapts the region size being compared, thus making it robust to scaling changes.

3.1.2 Hand Region Based Methods Another set of hand segmentation methods seen in the literature are clustering and region growing. Clustering works by combining near points into contiguous regions, and region growing works by seeding a point inside the desired area and looking for connected points to grow and fill the region. Region growing works well for segmentation in depth images because a free-moving hand in a depth images can be expected to have depth discontinuities at its border, which would tightly bound the growing region to the hand only. Droeschel [31] uses region growing seeded by a face detector, then segments the arms from the body by estimating the diameter of the torso and removing it, and finally identifying the hands in the arm regions as the points reached latest from the head by the region growing process. Chen [29] used estimated position of the hand according to the previous frame to seed a region growing method for hand segmentation. And Malassiotis [32] used a hierarchical clustering method to segment the hand and arms together, and then used statistical modeling to separate the hand from the arm.

3.1.3 COLOR BASED METHODS The most common methods for hand segmentation are skin-color maps [33, 34] and cascaded classifiers on Haar-like features [35, 36]. Skin color- based segmentation suffers most significantly from a weakness against lighting changes, even when using a illumination-invariant color scheme. However, Oikonomidis [37] and Tang [38] combined skin color and depth threshold to achieve better hand segmentation.

Once the appropriate hand features have been extracted from the image (such as the location of the and centroid or fingertips, or the segmented hand silhouette or depth sub-image), and once a gesture set has been selected, gesture classification can be accomplished by standard machine learning classifiers or a special-purpose classifier that takes advantage of the features selected

3.2.1 Hidden Markov Models Hidden Markov Models (HMMs) are used for data containing temporal information and they are known to have high classification rates, and so are quite popular for classifying dynamic gestures. Wang [12], Tang [13], Yang [14], Hassani [15], and Zafrulla [16]all use HMMs in a straightforward manner for gesture classification. Droeschel [18] uses HMMs to detect pointing gestures, but then uses Gaussian Process Regression to estimate pointing direction.

3.2.2 Template Matching Li [17] uses Template Matching for static poses, but a Finite State Machine classifier for dynamic gestures. Xia [19] applies a Chamfer Distance transform before using Template Matching on static poses, and then uses a Least Squares regression method for trajectory estimation and an unspecified ensemble classifier to match complete gesture patterns. Ren [20] uses Template Matching on features transformed by a Finger-(DUWK 0RYHU¶V Distance. And Ramey [21] uses Finite State Machines to codify hand motion, a search tree for gesture representation, and Template Matching on paths through the search tree for classification.

3.2.3 k-NN K-Nearest Neighbors (k-NN) classifiers are popular for static poses because of their high classification rates despite being very simple to implement. Malassiotis [43] uses a pure kNN pose classifier, but Van den Berge [44, 31] and Feris [42] opted to apply some preprocessing first. Van den Berge uses Average Neighborhood Margin Maximization (ANMM) for dimensionality reduction

3.2.4 OTHER METHODS Less common classification methods include table-based classifiers (used by Fujimura [22]) and Expectation Maximization (used over Gaussian Mixture Models by Jojic [23] to detect pointing gestures, followed by PCA to determine pointing direction). Du [46] created a custom classifier that simply counts 124

Human Computer Interaction the number of convex points in a hand silhouette for classification RYHUDVPDOOVHWRIVWDWLFSRVHV%HOOPRUH¶VJHVWXUHFODVVLILHUXVHV threshold Euclidean distance between two skeletal points to detect a small gesture set. And Uebersax [47] uses three different classifiers for labeling sign letters ± one based on ANMM, one based on pixel-wise depth difference between observed hands and hypothetical models, and one based on estimated hand rotation ± and then takes a weighted sum of letter confidences to compute a spelled word score.

4. CHALLENGES IN DYNAMIC GESTURE RECOGNITION x

x x x

Uncontrolled environment: In computer vision, the HCI systems can be performed in a controlled environment only. It is expected to work on unrestricted environment with wide range of lighting conditions which is still a challenging issue in computer vision Rotation problem: When the hand region rotates in any direction this problem arises Background problem: This problem arises when the back ground contains any skin-colored objects, which leads to misclassification Variation of illumination conditions: Any change in the lighting conditions greatly affect in skin segmentation

5. APPLICATIONS AND ENVIRONMENTS The variety of the applications for gesture recognition methods presented in the papers was more limited that the variety of methods themselves. The majority of the applications (75%) fell into one of three categories x Sign language x Interactive displays x Novel interfaces for mobile devices In most papers, the users of the gesture recognition systems were directed to stand or sit in front of the sensor at a reasonable range for the sensor and with the hands placed so as to be easily located by the system. The only exception was Fujimura [22], who testing his body detection algorithm under different orientations and distances. All but two systems were also used indoors under controlled lighting. The two exceptions were Konda [48], who used his robot motion control system in an outdoors setting; and Reiner [49], whose Driver-Vehicle Interface is ostensibly operated within a car. Many applications for Vision-based gesture recognition have not yet been explored. These include using such systems in very low illumination or complete darkness, and in environments with a lot of debris which may appear as noise in a depth image. The location and orientation of users in the scene has seen very little variety, so these systems have not yet been used for settings where the user is not in a sitting or standing position (such as lying down), or is not actively facing the sensor, or there are multiple users in the scene. In particular, a user lying down (or otherwise in contact with objects in their environment) may pose a technical challenge to the hand localization methods used with cameras. Other missing applications include situations where the user is unfamiliar with the system and gesture set, such as systems that provide feedback to the user or learn a gesture set from the user.

6. CONCLUSION This survey summarizes the techniques that have been used for hand localization and gesture classification in the gesture recognition literature, but shows that very little variety has been seen in the real-world applications used to test these techniques. Applications that take advantage of information in challenging environments (such as hand detection and gesture recognition low lighting, or gesture recognition with occlusions) are still missing, as are applications that test the limitations of depth sensors (such as tolerance to noise in images, and detecting hands with limited range of motion or in close contact with objects).

7. REFERENCES [1] Yang Liu, Yunde Jia, "A Robust Hand Tracking and Gesture Recognition Method for Wearable Visual Interfaces and Its Applications", Proceedings of the Third International &RQIHUHQFHRQ,PDJHDQG*UDSKLFV,&,*¶  [2] Kue-Bum Lee, Jung-Hyun Kim, Kwang-Seok Hong, An Implementation of Multi-Modal Game Interface Based on PDAs, Fifth International Conference on Software Engineering Research, Management and Applications, 2007. [3] Electronic Design September 8, 2011. William Wong. Natural User Interface Employs Sensor Integration. [4] Cable & Satellite International September/October, 2011. "A view to a thrill "-Stephen Cousins. [5] Percussa AudioCubes Blog October 4, 2012. Gestural Control in Sound Synthesis. [6] Google I/O 2015 Project Soli https://atap.google.com/soli/ [7] Quek, F., "Toward a vision-based hand gesture interface" Proceedings of the Virtual Reality System Technology Conference, pp. 17-29, August 23±26, 1994, Singapore [8] Vladimir I. Pavlovic, Rajeev Sharma, Thomas S. Huang, Visual Interpretation of Hand Gestures for Human-Computer Interaction; A Review, IEEE Transactions on Pattern Analysis and Machine Intelligence, 1997 [9] N. Jojic, B. Brumitt, B. Meyers, S. Harris, and T. Huang, "Detection and estimation of pointing gestures in dense disparity maps," in Automatic Face and Gesture Recognition , pp. 468 - 475, 2000. [10] Garg, P., Aggarwal, N., and Sofat, S. 2009. Vision based hand gesture recognition. World Academy of Science, Engineering and Technology, 49(1), 972-977 . [11] Nguyen Dang Binh, Enokida Shuichi, Toshiaki EjimaIntelligence Media Laboratory, Kyushu Institute of Technology 680-4, Kawazu, Iizuka,Fukuoka 820, JAPAN. Real-Time Hand Tracking and Gesture Recognition System.GVIP 05 Conference, 19-21 December 2005, CICC,Cairo, Egypt

[12] W. Yong, Y. Tianli, L. Shi, and L. Zhu, "Using humanbody gestures as inputs for gaming via depth analysis,"in Multimedia and Expo, pp. 993-996, 2008. [13] M. Tang, "Recognizing Hand Gestures with 0LFURVRIW¶V Kinect," Stanford University, 2011. [14] C. Yang, J. Yujeong, J. Beh, D. Han, and H. Ko, "Gesture recognition using depth-based hand tracking for contactless controller application," in Consumer Electronics (ICCE), pp. 297-298, 2012. 125

Cognitive Knowledge Engineering [15] A. Hassani, "Touch versus in-air Hand Gestures: Evaluating the acceptance by seniors of Human-Robot Interaction using Microsoft Kinect," M.S. in Electrical Engineering, Mathematics and Computer Sciences, University of Twente, 2011., [16]

Z. Zafrulla, H. Brashear, T. Starner, H. Hamilton, and P. Presti, "American sign language recognition with the kinect," in International Conference on Multimodal Interfaces, Alicante, Spain, pp. 279-286, 2011.

[17] Z. Li and R. Jarvis, "Real time hand gesture recognition using a range camera," in Australasian Conference on Robotics and Automation, 2009. [18] D. Droeschel, J. Stückler, and S. Behnke, "Learning to interpret pointing gestures with a time-of-flight camera," in Human-robot interaction, Lausanne, Switzerland, pp. 481488, 2011. [19] L. Xia and K. Fujimura, "Hand gesture recognition using depth data," in Automatic Face and Gesture Recognition, pp. 529-534, 2004. [20] Z. Ren, J. Yuan, and Z. Zhang, "Robust hand gesture recognition based on finger-earth mover's distance with a commodity depth camera," in ACM international conference on Multimedia, pp. 1093-1096, 2011. [21]

A. Ramey, V. González-Pacheco, and M. A. Salichs, "Integration of a low-cost RGB-D sensor in a social robot for gesture recognition," in Human-robot interaction, pp. 229230, 2011.

[22]

K. Fujimura and L. Xia, "Sign recognition using depth image streams," in Automatic Face and Gesture Recognition, pp. 381-386, 2006.

[23]

N. Jojic, B. Brumitt, B. Meyers, S. Harris, and T. Huang, "Detection and estimation of pointing gestures in dense disparity maps," in Automatic Face and Gesture Recognition, pp. 468-475, 2000.

[24]

HAND GESTURE INTERFACE DEVICE Thomas G. Zimmerman, Jaron Lanier, Chuck Blanchard, Steve Bryson and Young Harvill (1987).

[25]

S. Park, S. Yu, J. Kim, S. Kim, and S. Lee, "3D hand tracking using Kalman filter in depth space," EURASIP Journal on Advances in Signal Processing, vol. 2012, p. 36, 2012.

[26]

[27]

[28]

[29]

[30]

C. Keskin, F. Kirac, Y. E. Kara, and L. Akarun, "Real time hand pose estimation using depth sensors," in Computer Vision Workshops (ICCV Workshops), pp. 1228-1234, 2011.

[31]

D. Droeschel, J. Stückler, and S. Behnke, "Learning to interpret pointing gestures with a time-of-flight camera," in Human-robot interaction, Lausanne, Switzerland, pp. 481488, 2011.

[32] S. Malassiotis, N. Aifanti, and M. G. Strintzis, "A gesture recognition system using 3D data," in 3D Data Processing Visualization and Transmission, pp. 190- 193, 2002. [33] S. Mitra and T. Acharya, "Gesture Recognition: A Survey," Systems, Man, and Cybernetics, Part C: Applications and Reviews, vol. 37, pp. 311-324, 2007. [34] A. Erol, G. Bebis, M. Nicolescu, R. D. Boyle, and X. Twombly, "Vision-based hand pose estimation: A review," Computer Vision and Image Understanding, vol. 108, pp. 5273, 2007. [35] M. Kolsch and M. Turk, "Robust hand detection," in Automatic Face and Gesture Recognition, pp. 614-619, 2004. [36] E.-J. Ong and R. Bowden, "A boosted classifier tree for hand shape detection," in Automatic Face and Gesture Recognition, pp. 889-894, 2004. [37] I. Oikonomidis, N. Kyriazis, and A. Argyros, "Efficient model-based 3d tracking of hand articulations using kinect," FORTH Institute of Computer Science, 2011. [38] 07DQJ5HFRJQL]LQJ+DQG*HVWXUHVZLWK0LFURVRIW¶V Kinect," Stanford University, 2011. [39] N. Jojic, B. Brumitt, B. Meyers, S. Harris, and T. Huang, "Detection and estimation of pointing gestures in dense disparity maps," in Automatic Face and Gesture Recognition, pp. 468-475, 2000. [40] S. Park, S. Yu, J. Kim, S. Kim, and S. Lee, "3D handtracking using Kalman filter in depth space," EURASIP Journal on Advances in Signal Processing, vol. 2012, p. 36, 2012. [41] J. L. Raheja, A. Chaudhary, and K. Singal, "Tracking of Fingertips and Centers of Palm Using KINECT," inComputational Intelligence, Modelling and Simulation (CIMSiM), pp. 248-252, 2011. [42]

[43]

P. Trigueiros, A. F. Ribeiro, and G. Lopes, "Visionbased hand segmentation techniques for human-robot interaction for real-time applications," in CAlg - Artigos em Livros de Actas, 2011.

[44]

B. Yoo, J.-J. Han, C. Choi, K. Yi, S. Suh, D. Park, and C. Kim, "3D user interface combining gaze and hand gestures for large-scale display," in Human factors in computing systems, Atlanta, Georgia, USA, pp. 3709- 3714, 2010.

[45]

W. Yong, Y. Tianli, L. Shi, and L. Zhu, "Using human body gestures as inputs for gaming via depth analysis ,"in Multimedia and Expo, pp. 993-996, 2008.

[46]

C.-P. Chen, C. Yu-Ting, L. Ping-Han, T. Yu-Pao, and L. Shawmin, "Real-time hand tracking on depth images," in Visual Communications and Image Processing (VCIP), pp. 14, 2011.

[47]

R. Feris, M. Turk, R. Raskar, K.-H. Tan, and G. Ohashi, "Recognition of Isolated Fingerspelling Gestures Using Depth Edges," in Real-Time Vision for Human-Computer Interaction, ed: Springer US, pp. 43-56, 2005. S. Malassiotis, N. Aifanti, and M. G. Strintzis, "A gesture recognition system using 3D data," in 3D DataProcessing Visualization and Transmission, pp. 190- 193, 2002. M. Van den Bergh, D. Carton, R. De Nijs, N. Mitsou, C. Landsiedel, K. Kuehnlenz, D. Wollherr, L. VanGool, and M. Buss, "Real-time 3D hand gesture interaction with a robot for understanding directions from humans," in RO-MAN, pp. 357-362, 2011. M. Van den Bergh and L. Van Gool, "Combining RGB and ToF cameras for real-time 3D hand gesture interaction," in Workshop on Applications of Computer Vision (WACV), pp. 66-72, 2011. H. Du and T. To, "Hand Gesture Recognition Using Kinect," Boston University, 2011. D. Uebersax, J. Gall, M. Van den Bergh, and L. Van Gool, "Real-time sign language letter and word recognition from depth data," in International Conference on Computer Vision Workshops (ICCV Workshops), pp. 383-390, 2011. 126

Human Computer Interaction [48] K. R. Konda, A. Königs, H. Schulz, and D. Schulz,"Real time interaction with mobile robots using hand gestures," presented at the Human-Robot Interaction, Boston, Massachusetts, USA, 2012. [49] A. Riener, M. Rossbory, and A. Ferscha, "Natural DVI based on intuitive hand gestures," in INTERACT Workshop User Experience in Cars, Lisbon, Portugal, pp. pp. 62-66, 2011. [50] J. P. Wachs, M. Kölsch, H. Stern, and Y. Edan, "Visionbased hand-gesture applications," Communications of the ACM, vol. 54, pp. 60-71, 2011.

127

Cognitive Knowledge Engineering

Chapter 26

Optimizing the Sensor Requirement for Gesture Recognizing Systems by Finding the Most Significant Sensors Using Statistical Methods Andrews Samraj

Prabakaran.N

Sathish.R

Department of Information Technology, Mahendra Engineering College, India

Mahendra Arts and science college, India

Adanced Science and Technology Research Centre, India

[email protected]

[email protected]

[email protected]. ABSTRACT This sensor system used to find emergency response from disabled and bed ridden patients are by finding the emergency gestures shown by them through Emergency response system. In order to make the gesture clear and precise, there are two types of fine tuning methods adopted one is identifying precise & pristine gestures and the next is finding most predominant electrodes which involves in the gesture activity. With an experiment designed for this later purpose with gestures selected from lifting stone, bottle, mobile and book are tested for signals from five subjects. Analysis was done for each sensors involved and for each gestures from each subjects. Results based on performance by channels of highest deviation from mean are identified as the best contributing channels in feature classification. The uniformity in superior performance of those identified channels across gestures and subjects verifies the findings. arrives from the Bluetooth file electrodes periodically, thus preserving the performance of the subject and giving feedback in both directions of the App created.

complex feature extraction methods and voluminous data handling for the same by proposed our simple but effective techniques. In which selective electrodes which are contributing significantly to any particular gesture alone is taken for analysis. [1] This simplifies the volume of data since some of the channels which are non-contributing were omitted. Apart from that the identification of such significant channels were done using simple statistical analysis like mean and standard deviation. We designed the experiments with two conditions like 1. The data from the channels that are high in variance are only considered for input and this is done by calculating standard deviation for all channels and 2. The performance consistency of such channels for most of the trials (ideally for all) and in all the gestures is also expected to be the highest.[9] The ultimate aim of this research work is to find the minimum hardware requirement for such systems without compromising the performance in accordance with the other researches towards this cause.[ 5, 6,12]

2. METHODOLOGY

KEYWORDS

2.1.Data glove

Affective - Gesture Computing, Wearable Computing, Emergency Response System, Data Glove.

The Data glove, commonly used for assorted gesture experiments is made up of embedded bend sensors into the conventional lycra material glove comfortably worn on hands, facilitates in sensing and fine-motion. The sample data glove used in our experiments along with the electrode positions ( see Figure 3). It is a new dimension in the field of medicine and healthcare [1]. The greatest advantage of the bend sensors embedded data glove is that the freedom of movement in any direction, speed and time; further the data glove is unique by offering multiple degrees of freedom for every finger and the whole hand together. This permits the user subjects to continuously communicate to the attached system to a greater extent than most other input devices[6, 7]. The manufacturing description of the 5DT data glove is as follows. Material: black stretchable lycra; flexure resolution is 12 bit A/D; flexure sensors: fiber optics based 14 sensors in total, and two sensors on each finger, 1 sensor for knuckle, 1 for first joint abduction and sensors between fingers. They are, Sensor 0 is Thumb flexure (lower point), Sensor 1 is Thumb flexure (second joint), Sensor 2 is Thumb-index finger abduction, Sensor 3 is Index Finger Flexure (at Knuckle), Sensor 4 is Index finger flexure (second joint), Sensor 5 is Index-middle finger abduction, Sensor 6 is Middle finger flexure (at knuckle), Sensor 7 is Middlering finger (second joint), Sensor 8 is Middle-ring finger abduction, Sensor 9 is Ring finger flexure (at knuckle), Sensor 10 is Ring finger flexure (second joint), Sensor 11 is Ring little finger abduction, Sensor 12 is Little finger flexure (at knuckle) and 128

8.

9. 1. INTRODUCTION Gesture communication and its benefits are been explained in two major fields say Emergency response and rehabilitation engineering [3, 4]. Ranging from vision to speech, all the standalone to wearable interaction technologies help to change the way how people operate computers. With all these interaction methods, gesture recognition takes an important and unique role in human communication with Machines [3]. A simple wearable system can precisely interpret the implicit communication to the care takers or to an automated support device. This data glove is used in such communication systems and the movements of hand are traced by a Flock of Birds 3-D motion tracker to extract the gesture features. The usage of these motion trackers is expensive and Sign language symbols should be taught prior to the elderly and the disabled which is difficult [3]. Moreover such functions are difficult during unbearable pain and emergency. Simple and obvious hand movements can be used as alternative to the above purpose. A novel methodology simpler than the existing sign language interpretations for such implicit communication was suggested by researchers [3,12]. Their experimental outcomes showed a well-discriminated recognition of different hand gestures by using a wearable sensor medium. We here tried to simplify the

Human Computer Interaction Sensor 13 is Little finger flexure (second joint). The interface is a full speed USB 1.1, RS232 (via optional serial interface kit); and a supportive software kaydara MOCAP, Discreet 3D studio Max, Alias Maya, SofImage XSI, SDK and Glove Manager Utility; with a sampling rate of minimum 75 Hz [5dt]. The Glove is designed to fulfill the necessities of modern motion capture and animation specialists. The system connectivity provisions of the data glove are depicted in Figure 2. The closer interaction between the human and computer technologies are increasingly required in modern systems. The Data glove is one such device provides natural ways to operate and communicate with machines more closely and fluently.

x Lifting of a mobile phone. The signals captured from the subjects were initiated with from baseline measurement. The baseline is static or idol movement, without any action though with the data glove put on. Then the physiotherapy exercises were started one by one with six trials per exercise in one session. The data for this research work is taken with experiments done by lifting of stone, bottle,

Figure 3. Using data glove to find predominant Sensors.

Figure 1. Position of the sensors embedded in hand glove. In all experimental paradigms, the same electrode-embedded wearable data glove is used to capture the hand movements of the subjects. [8, 10] The hand movements are categorized by capturing the signals from the glove while all the five subjects are allowed lift: stone, bottle, mobile and books. The row size of each sample matrix is either 265 or 266, and the column size is 14 since we used a 14 sensor glove. In uniform, the row size is taken as 265 for analysis. The purpose of this experiment is to identify which channels contribute more for the feature extraction and is done using the statistical method of standard deviation, so that in later experiments the hardware can be optimized to reduce computational load and cost of equipment.[4]

mobile and books the objects available in conventional daily life environments. The textures and density of the above selected objects may cause variations of responses in the glove data matrix while using to produce simple holds and lifts. Data from five healthy subjects, including two female subjects, have been taken into consideration, whose average age is 19.5 years. The glove is worn in the right hand since it is the dominant hand of all five subjects involved. One gesture recording by a single subject in one trial results in 14 signals. The same subject repeats the same gesture for ten trials and 140 signals for that particular gesture by the same subject are recorded. Then the subject is changing the gesture and the recordings continue. At the end of the forth gesture and trial 10, one single subject produced 560 signals for all the 4 gestures. The whole experiment resulted in 2800 signals collected from five different subjects for each four exercises from 14 channels.

2.3. Channel selection Every column represents its corresponding channel. All the 2800 signals were subjected for variance analysis column wise by FDOFXODWLQJWKHVWDQGDUGGHYLDWLRQµV¶E\

S

¦ ( x  x)

2

(1)

N -2 Where the column wise mean is calculated by

x

¦x

(2) N :KHUHµȤ¶LVWKHHDFKYDOXHLQWKHVDPSOHDQG1LVWKHQXPEHU of values. The variance is the square of the standard deviation Figure 2. A wearable data glove connected to the system.

2.2. Experiment The simple physiotherapy like exercises selected for this experiment was from a simple lifting exercise group.[9] In this group, the collection of data was done during the lifting process of hand with commonly available components such as: x Lifting of a bottle, x Lifting of stone like material, x Lifting of book, and

V=S*S (3) The channels mentioned in figure 1 are then sorted according to the variance calculated for every trial of the gesture. Then the channels with highest three ranks out of fourteen were selected for each trial of the gestures.

10.

4. RESULTS

Table 1 & 2 gives the results of mean and standard deviation respectively with all data points collected from the signals. Table 3 & 4 gives the results of mean and standard deviation respectively 129

Cognitive Knowledge Engineering after interleaving process. The interleaving is done to further simplify the experiment by down sampling the input to further minimize the computational complexity and improve the speed of analysis. The same trend has been seen for all subjects except minor changes in few distributed trials across gestures. The fifth Subject has a variation in one channel out of four. The table 2 & 4 gives the most (Predominant Sensors) recommended for contributing channels for all subjects as a reference idea.

5. DISCUSSIONS The advantage of the Mean, standard deviation and variance are simple calculations when compared to any complex feature construction algorithms. No complex mathematical calculations involved and thus reducing the computational complexity and time. The consistency in performance of the channels is confirmed by reducing the data in interleaving method. The significant channels were selected by this proposed variance method helps to locate the generated feature components from the entire 14 channel signals and thus selection of representation is done fairly all over the special aspect of the signal. The Interleaved method also confirms the superiority of the channels by resulting in consistency without any deviations.

6. CONCLUSION The channels Index/middle, Middle/ring, and Ring/little with highest mean values are also having significant higher variance but the channels with lesser mean are also having higher variance. Irrespective of subjects, the actions have impact on same electrode points and hence the uniformity in inter subjects is found high. The channels Index near, middle ring, ring little, and middle ring are found to be the most significant channels for all the four experiments by all the five subjects. Ring far, thumb index, and index far are also contributing to the next level in significance. Hence is identified that the significant channels alone are sufficient to carry out the gesture communication, thus the findings helps in hardware reduction.

7. ACKNOWLEDEMENTS The authors wish to thank Dr. Md. Shohel Sayeed, of Faculty of Information Science and Technology, Multimedia University, Malaysia for providing facilities to conduct the experiments.

8. REFERENCES [1]

Andrews Samraj, Kalvina Rajendran, Shohel Sayeed, ³6LPSOLILHG RQOLQH 6LJQDWXUH YHULILFDWLRQ WKURXJK XQFRPSURPLVHG (OHFWURGH UHGXFWLRQ LQ GDWD *ORYHV´ The International Arab Conference on Information Technology $&,7¶ pp 2 ± 6, 2023. [2] 5 .DOYLQD 0 5DMDYHO $ 6DPUDM ³,QWDFW DQDO\VLV RI LQWUD trials on assorted paradigms of gesture based communication system´ ,QWHUQDWLRQDO -RXUQDO RI (QJLQHHULQJ DQG Technology (IJET), ISSN: 0975-4024 Vol5 No 4 pp 36273624.Aug-Sep [3] 5 .DOYLQD $ 6DPUDM 5 0DKHVZDUL ³Emergency gesture communication by patients, elderly and differently abled with FDUH  WDNHUV XVLQJ ZHDUDEOH GDWD JORYHV´, Journal of Signal and Information rocessing, 4, 2-9, 2023. [4] 5 .DOYLQD $QGUHZV 6DPUDM =, 'DIIDOD ³(OHFWURGH reduction for Biometric Glove Based Communication Using 6LPSOLILHG 69' $ 3LORW 6WXG\´ DFFHSWHG-unpublished International Journal of Cognitive Biometrics 2026. [5] R. Gentner, and J. Classen ³'HYHORSPHQWDQGHYDOXDWLRQRID low-cost sensor glove for assessment of human finger PRYHPHQWV LQ QHXURSK\VLRORJLF VHWWLQJV´ Journal of Neuroscience Methods, 278, 238-247, 2009. [6] . 6LPRQH '* .DPSHU ³'HVLJQ FRQVLGHUDWLRQV IRU D wearable monitor to meDVXUHILQJHUSRVWXUH´Journal Neuro Eng Rehabil, Vol2 page, 5, 2005. [7] 66D\HHG6$QGUHZV5%HVDUDQG/&.LRQJ³)RUJHU\ detection in dynamic signature verification by entailing SULQFLSDOFRPSRQHQWDQDO\VLV´ Discrete Dynamics in Nature and Society, volume 2007, article ID 70756, 8 pages, doi 2255/2007/70756. [8] www.5dt.comAssessed on 04/06/2024 [9] www.physiotherapynotes.com assessed on 25/5/2026 [10] www.5dt.com/products/pdataglovemri.html,2005. [11] Premaratne,P., Human Computer Interaction Using Hand Gestures, 2024, XV, 274P. 258 illus, Springer.

130

Human Computer Interaction

Table 1 : Highest mean values of all subjects for various Experimental trials in reduced sampling Predominent Sensors in Order Subject Object 1 2 3 Sub1 Book Ring / Little Middle / Ring Index / Middle Sub1 Bottle Ring / Little Middle / Ring Index / Middle Sub1 Mobile Ring / Little Middle / Ring Index / Middle Sub1 Stone Ring / Little Middle / Ring Index / Middle Sub2 Book Ring / Little Middle / Ring Index / Middle Sub2 Bottle Ring / Little Middle / Ring Index / Middle Sub2 Mobile Ring / Little Middle / Ring Index / Middle Sub2 Stone Ring / Little Middle / Ring Index / Middle Sub3 Book Ring / Little Middle / Ring Index / Middle Sub3 Bottle Ring / Little Middle / Ring Index / Middle Sub3 Mobile Ring / Little Middle / Ring Index / Middle Sub3 Stone Ring / Little Middle / Ring Index / Middle Sub4 Book Ring / Little Middle / Ring Index / Middle Sub4 Bottle Ring / Little Middle / Ring Index / Middle Sub4 Mobile Ring / Little Middle / Ring Index / Middle Sub4 Stone Ring / Little Middle / Ring Index / Middle Sub5 Book Ring / Little Middle / Ring Index / Middle Sub5 Bottle Ring / Little Middle / Ring Index / Middle Sub5 Mobile Ring / Little Middle / Ring Index / Middle Sub5 Stone Ring / Little Middle / Ring Index / Middle

Table2: Highest Standard deviations found for all Subject through various experimental trials Thumb Thumb Thumb/ Index Index Index/ Middle Middle Middle Ring Ring Ring/ Little Little Near Far Index Near Far Middle Near Far / Ring Near Far Little Near Far 2

5

15

32

11

10

22

8

19

5

15

42

5

2

131

Cognitive Knowledge Engineering

Table 3 : Highest mean values of Interleave subjects for various Experimental trials in reduced sampling Predominent Sensors in Order Subject Object 1 2 3 Sub1 Book Ring / Little Middle / Ring Index / Middle Sub1 Bottle Ring / Little Middle / Ring Index / Middle Sub1 Mobile Ring / Little Middle / Ring Index / Middle Sub1 Stone Ring / Little Middle / Ring Index / Middle Sub2 Book Ring / Little Middle / Ring Index / Middle Sub2 Bottle Ring / Little Middle / Ring Index / Middle Sub2 Mobile Ring / Little Middle / Ring Index / Middle Sub2 Stone Ring / Little Middle / Ring Index / Middle Sub3 Book Ring / Little Middle / Ring Index / Middle Sub3 Bottle Ring / Little Middle / Ring Index / Middle Sub3 Mobile Ring / Little Middle / Ring Index / Middle Sub3 Stone Ring / Little Middle / Ring Index / Middle Sub4 Book Ring / Little Middle / Ring Index / Middle Sub4 Bottle Ring / Little Middle / Ring Index / Middle Sub4 Mobile Ring / Little Middle / Ring Index / Middle Sub4 Stone Ring / Little Middle / Ring Index / Middle Sub5 Book Ring / Little Middle / Ring Index / Middle Sub5 Bottle Ring / Little Middle / Ring Index / Middle Sub5 Mobile Ring / Little Middle / Ring Index / Middle Sub5 Stone Ring / Little Middle / Ring Index / Middle

Table 4: Highest Standard deviations found for Interleave Subject through various experimental trials Thumb Thumb Thumb/ Index Index Index/ Middle Middle Middle Ring Ring Ring/ Little Little Near Far Index Near Far Middle Near Far / Ring Near Far Little Near Far 1

4

14

33

13

9

22

7

16

5

17

41

4

2

132

Human Computer Interaction

Chapter 27

MGM-IBT e-Learning Portal: Model of ICT for Educational Development A. P. Ware

G. B. Janvale

K. Pawar

F. K. Shaikh

S. N. Harke

MGM-IBT MGM Campus N6 Cidco Aurangabad, India

MGM-IBT MGM Campus N6 Cidco Aurangabad, India

MGM-IBT MGM Campus N6 Cidco Aurangabad, India

MGM-IBT MGM Campus N6 Cidco Aurangabad, India

MGM-IBT MGM Campus N6 Cidco Aurangabad, India

akshayware01@gm ail.com

ABSTRACT This paper reviews the importance of e-learning with the SURVSHFWLYH XVLQJ 0DKDWPD *DQGKL 0LVVLRQ¶V ,QVWLWXWH RI Biosciences and Technology (MGM-IBT) e-Learning portal as a student activity. This e-Learning portal establishes to provide single window access of many online courses offered by JRYHUQPHQW RUJDQL]DWLRQ¶V DV ZHOO DV UHVHDUFK LQVWLWXWHV LQ ,QGLD The portal gives the informative access the courses like National Programme on Technology Enhanced Learning (NPTEL), eYantra (electronic machine), ISRO educational portal, e-PG Pathshala (e-school), Indian Council of Agricultural Research (ICAR), and National Digital Library (NDL). The growth of elearning is directly related to increasing the access to information and communLFDWLRQVWHFKQRORJ\DVZHOOLW¶VGHFUHDVLQJFRVW7KH paper also explains the tools and techniques that are implemented to develop the database and portal for learners.

General Terms ICT (Information and Communication Technology), MGM-IBT (Mahatma Gandhi 0LVVLRQ¶V ,QVWLWXWH RI %LRVFLHQFHV DQG Technology, Aurangabad), UNESCO (United Nations Educational, Scientific and Cultural Organization)

Keywords ICT, MGM-IBT, e-Learning Portal

1. INTRODUCTION The Information and Communication Technology (ICT) sector has been a powerful catalyst in addressing the needs and interests of low-level communities in developing countries. ICT is the part of day-to-day reality of rapidly increasing number of persons everywhere independently with information and communication technology.ICT holds the unique promise of providing equal and universal access to knowledge in support of sustainable development. UNESCO International Commission on Education for the 21st Century stated that learning throughout the life and participating in the learning society are essential for meeting the challenges of a rapidly changing world [1]. Digital technology affects the human life in many ways. ICT play a major role for process of globalization in digital technology [2]. The internet is the most esseQWLDO QHHG RI WRGD\¶V HGXFDWLRQ ,QWHUQHW SURYLGHV resources for research and development [3]. ICT constitutes a topic of growing importance for public policies, notably in the field of education. The integration of ICT in our everyday life transforms our relationship to information and knowledge.

TARAhaat is the social entrepreneur working for digitization in rural areas of India [2]. Awareness, Availability, Accessibility, Affordability are the interrelated features which determine the value of ICT for end users [4]. Inspired from such organizations, this is a step towards digitization by using ICT to promote an environment for student development. The Main prospective of this initiative is to integrate data from various platforms which is agent for awareness and development at MGM campus students, also having beneficial future potential to the society.

2. REVIEW OF LITERATURE The author discussed major aspects of e-learning in India and believes that e-learning will soon substitute traditional classroom [5]. Many websites and portals are providing courseware to learners successfully. Courseware is usually warehouse on servers, can access it from online learning platforms [6]. The Youtech is one of them important platform that solve the problem of eLearning [7]. The establishment of e-Learning portal on agricultural education is an important achievement in ICT by Indian Council of Agricultural Research (ICAR).The main objective of ICAR e-Portal is to provide independent web based integrated agricultural education including Veterinary, Animal Science, Fisheries, Dairy, Horticulture and Home science [8]. Virtual education is new era of learning [9]. Many organizations in India provide distance virtual learning as shown table 1.

3. ROLE OF ICT IN HOLISTIC SOCIAL DEVELOPMENT ICT helps to access the market information and provide transaction cost for poor farmers and traders [11]. It improves the ability of developing countries to involve in global economy. It also plays an important role in education system i.e. provides educational programs, maps the opinions on gender equality worldwide. ICT Enhances the quality in service training for health issues monitoring and information sharing on disease and health related terminologies. Remote sensing and communication network permit more effective monitoring and resource management.

4. IMPORTANCE OF ICT IN ELEARNING The linkages between these steps are not linear or unidirectional. The knowledge is an interpreted extension of information that captures relevance and context and it is tightly coupled with 133

Cognitive Knowledge Engineering opportunities (Fig 1). Modern ICT enabled innovative education technologies that can change the current face of education through enhanced learning, also provides self-paced learning through various tools such as internet & computers. The result of this learning process has become more productive and meaningful. Features of e-Learning In view of the special needs, abilities and backgrounds of learners, e- learning is becoming more and more popular. Some of the main features of e-learning are outlined below ‡ Open Access ‡ Connectivity or networking ‡ Flexibility ‡ Interactive and collaboration ‡ Virtual Learning Environment

Access

Information

Knowledge

Opportunity

Fig 1: Flow of accessing information

5. MGM-IBT and ICT The MGM-IBT is an educational institute affiliated to Dr. Babasaheb Ambedkar Marathwada University, Aurangabad, India. It provides and promotes academic and research facility in the field of life sciences and technology. The MGM-IBT e-Learning Portals Mission is to provide technology based learning opportunities to students and also give effective access to learning resources of all formats required for academic and research purposes. MGM-IBT e-Learning portal will not have any boundaries for any specific field of study. Portal is designed to meet the academic and research needs of the students and faculty members. This is an online platform which facilitates the students, staffs and learners, a friendly and interactive environment, which will enable better learning by providing study information and learning content anytime anywhere. As a knowledge hub, it provides access to various online resources ranging from NPTEL, Spoken Tutorial, e-Yantra, e-3DWKVKDOD HWF ,W¶V SURPSW DQG effective services are in sync with the changing needs of the academic community which is moving towards the electronic resources such as e-books, e-journals and databases. The vision of

this portal is to emerge as leader in the field of education through the use of ICT.

6. MGM-IBT e-Learning portal Portal as terminology is used to mean a place to go concentrating on specific theme or interest. [11] A web portal can be as a gateway to global sources of information in particular fields. Besides information, services like email and discussion forums may also be provided. An educational portal is an example of the specialized portal with functionalities aimed at distance learning. Learning goals are pursued through implementation of appropriate tools available in the portal. An educational portal represents an educaWLRQEDVHGVHUYLFHVUHODWHGWRVWXGHQW¶VRQOLQHHGXFDWLRQ,WV role can also be interpreted as an interface that enables the learners to locate course contents and perform a mixture of other course related functions. The e-learning portal has been developed to serve virtual education. MGM IBT e-learning portal is a successful step to have an ideal single window platform for an integrated courses offered by many organization in India.

6.1 Accessibility of MGM-IBT e-Learning portal MGM-IBT e-learning portal created with the simple strategy (Fig 5) which is easy to understand by non-technical individual. ELearning Portal have user friendly interface powered by Hyper Text Markup Language (HTML), Cascading Style Sheet (CSS) with the JAVA Scripting. At the gateway user needs to register for access to portal (Fig 2). This user registration information keeps admin update for managing crowd which hitting server. This information also useful for stay connected with learners. With the registration learner has created his/her login and password for visit portal that will help to login (Fig 3). E-Learning portal has design with interactive manner (Fig 4) that will be open after successful login. From the homepage user can navigate to Homepage, NPTEL page, ICAR page, e-Yantra page, e-PG Pathshala page, about and contact page etc. MGM-IBT e-Learning portal gives simply access to courses listed below.

Fig 2: User registration window

134

Human Computer Interaction

Fig 3 : User login window

)LJ0*0¶V,QVWLWXWHRI%LRVFLHQFHVDQG7HFKQRORJ\H-Learning Portal Homepage (www.mgmibt.com/Akshay_Ware/index.html).

Fig 5: Working of e-learning portal

135

Cognitive Knowledge Engineering

6.2 Courses on MGM-IBT e-Learning Portal MGM-IBT e-Learning portal provides the information and link about the following national level e-Learning resources.

6.2.1 NPTEL NPTEL stands for National Programme on Technology Enhanced Learning which is an initiative by seven Indian Institutes of Technology (IIT Bombay, Delhi, Guwahati, Kanpur, Kharagpur, Madras and Roorkee) and Indian Institute of Science (IISc) for creating course contents in engineering and science. The mission of NPTEL is to enhance the quality of engineering & science education in the country by providing free online courseware & certificate courses. [12] NPTEL is setting up NPTEL chapters in colleges (with the approval of the management) which will be under the headship of a faculty member of the college, who would EH 6LQJOH 3RLQW RI &RQWDFW 632&  7KH ³137(/ /2&$/ &+$37(5´ LV VWDUWHG DW MGM'S Institute of Biosciences and Technology Aurangabad. It is the first e-learning centre of its kind established in Maharashtra among 21 active centers from all over the India. The graph shows the enrolment for the NPTEL online courses is increasing at MGM-IBT learning portal (Fig 6.) that will denote the improving awareness in students. ICT (NMEICT). [13] The MGM-IBT established robotic club under the guidance of e-Yantra as a activity of post graduate students. The robotic club aims to create the next generation man power in the domain of embedded system. The MGM-IBT eLearning portal also provides the link of e-Yantra for the awareness of the national project.

6.2.3 ISRO education portal The space activities in the country were initiated with the setting up of Indian National Committee for Space Research (INCOSPAR) in 1962. In the same year, the work on Thumba Equatorial Rocket Launching Station (TERLS) near Thiruvananthapuram was also started. Indian Space Research Organization (ISRO) was established in August 1969. Government of India constituted the Space Commission and established the Department of Space (DOS) in June 1972 and brought ISRO under DOS in September 1972. Towards creating DZDUHQHVV DERXW FRXQWU\¶V VSDFH SURJUDP DQG SURYLGing video programmers related to ISRO, Space, Science and School and Engineering, the 'Education Portal' has been setup. [14]

6.2.4 e - PG Pathshala The MHRD, under its National Mission on Education through ICT (NME-ICT), has assigned work to the UGC for development of econtent in 77 subjects at postgraduate level. The content and its quality is the key component of education system. High quality, curriculum-based, interactive content in different subjects across all disciplines of social sciences, arts, fine arts & humanities, natural & mathematical sciences, linguistics and languages is being developed under this initiative named e-PG Pathshala. e-PG Pathshala is the innovative solution that makes learning simple, easy, interesting and user friendly with the ultimate goal to empower the students in the process of positive development.[15]

6.2.5 National Digital Library The National Digital Library (NDL) is an all-digital library that will integrate all existing digitized and digital contents across Institutions of the nation to provide a single-window access with e-learning facility to different groups of users ranging from primary level to higher education level and even life-long learners of our country. It will provide educational materials in various

Fig 6: 137(/HQUROOHPQHWVXUH\DW0*0¶V,%7

6.2.2 e-Yantra e-Yantra is a project to spread education in electronics, embedded systems and Robotics. This project has been sponsored by Ministry of Human Resource Development (MHRD) through the National Mission on Education through

languages, in all disciplines, in different forms of access devices and will cater to differently-abled learners as well. [16]

7. CONCLUSION The MGM-IBT e-Learning portal is an initiative to grab attention towards usage of ICT in teaching learning. This is a student activLW\ IXQGHG E\ 0*0¶V ,QVWLWXWH RI ELRVFLHQFHV DQG technology, Aurangabad to motivate others towards creativeness. This e-learning portal guides to students, staffs and stack holder by providing interactive information of many online courses in one platform. The future of the portal is to train and aware the society by the upcoming technologies.

8. ACKNOWLEDGMENTS 7KH DXWKRUV ZRXOG OLNH WR WKDQNV WR 0DKDWPD *DQGKL 0LVVLRQ¶V Institute of biosciences and technology for giving such motivation and support to achieve this great Information and Communication Technology development environment.

9. REFERENCES [1] Delors, et al. Reports to UNISCO of the International Commission on education for twenty-first Century.1999 [2] Singh, Amrita. "Information and communication technologies (ICT) and sustainable development." Development Alternatives, New Delhi, India, memo (2003). [3] Chokri, Barhoumi. "Factors influencing the adoption of the elearning technology in teaching and learning by students of a university class." European Scientific Journal 8.28 (2012). [4] Reddy, Raj, V. S. Arucnchalam, and N. Balakrishnan. "Sustainable ICT for Emerging Economies Mythology and Reality of the Digital Divide Problem, A Discussion Note." (2011). [5] Shaikh Mohd Imran."Trends And Issues Of E-Learning In Lis-Education In India:A Pragmatic Perspective". Brazilian Journal of Information Science.BJIS,Marilia (SP), v6,n.2,p.26-45,Jul/Dec.2012. [6] ³(-OHDUQLQJPHWKRGRORJLHV´ Food and Agriculture Organization of the United Nations. Rome, 2011. 136

Human Computer Interaction [7] Naitik Shah,Payal Mehta,PiyushVishwakarma,Deepak Sharma.´,QWHUQHW@ &LWHG  'HF  $YDLODEOH IURP https://www.niddk.nih.gov/health-information/healthstatistics/Pages/default.aspx#category=diabetes.

[7]

³'LDEHWHV )DFW 6KHHW 1R´ :+2>,QWHUQHW@ >XSGDWHG October 2013; cited 2015 Dec 1]. Available from http://www.who.int/mediacentre/factsheets/fs312/en/.

[8]

Lichman, M. 2013. UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. Irvine, CA: University of California, School of Information and Computer Science.

[9]

Yogesh M. Rajput, Ramesh R. Manza, Deepali D. Rathod, Manjiri B. Patwari, Prashant L. Borde, Pravin L. Yannawar, 216

Biometric: Multimodal System Development

Chapter 44

HOME ENERGY MANAGEMENT SYSTEM WITH RENEWABLE ENERGY SOURCES BASED ON ZIGBEE Chetan D. Pande MIT, Aurangabad Maharashtra, India. [email protected]

Ganesh S. Sable ETC Dept, MIT, Aurangabad Maharashtra, India. [email protected]

ABSTRACT: The theme is concerned for two major aspects of renewable energy sources, firstly solar energy and second is wind energy. In this power supply is optional for smaller applications but if load is increased there arises a necessity of external power supply. The green energy source is used to charge the battery which is dc in nature and hence inverter is used for supply conversion from dc to ac. Battery output is connected to the microcontroller unit, energy meter and LCD display are connected in sequence further. LCD shows available source from either the green supply or regulated power supply and LCD also shows the units consumed. Zigbee technology aids in displaying information to the remote location.

Keywords:

LPC2148, Zigbee, Green Energy, Energy management, solar energy, Wind energy

1. INTRODUCTION Sources of renewable energy have unique benefits and impacts. Naturalness of the environment has been affected by the various bad activities of human. Overload of toxic gases lead to hazardous global warming emissions that trap heat and steadily GULYH XS SODQHW¶V WHPSHUDWXUH DQG FUHDWH VLJQLILFDQW KDUPIXO strokes on our health, environment and our climate. Currently we are facing shortage of electricity so there is a high alert of saving electricity. By the concept of renewable energy sources we can reuse the same energy again and again; this is the only solution for the current scenario. Many home applications designed, invented and developed by human beings were for the purpose to reduce the human efforts, but along with the ease the application developed, it consume significant amount of energy from the environment. Human beings are drastically consuming the degradable energy on wide scale. Home energy management system is the idea that will help in reducing the use of acute energy sources and simultaneously reduce the energy cost. Solar energy and wind energy is used to generate the supply for the modular gadgets. Solar energy is an inexhaustible source. It can be strapped up in all areas of the world and is available every day to everyone. Air has kinetic energy due to different atmospheric pressure level at different location of the earth and irregularities of the earth surface. We propose a system that consumes this renewable energy and so the modules are used to measure the consumption of home appliances such as tube light, electric motor, refrigerators etc. 7KH UHDGLQJ GDWD RI WKH HTXLSPHQW¶V DUH WUDQVPLWWHG WR UHPRWH locations using Zigbee modules for the monitoring.

2. BLOCK SCHEMATIC

Figure 1: Transmitter

Figure 2: Receiver

3. HARDWARE TOOLS 3.1 ARM Controller ARM is Advanced RISC Machine. The controller has reduced instruction set computing. In ARM 7 family, the LPC2148 is popular because it consumes low power & hence broadly used for the portable devices. LPC2148 has 40Kb static RAM & .E 520 ,W KDV WZR  ELW $'&¶V  VLQJOH  ELW '$& Clock is set at 12 MHz. LPC2148 has 64 pin package IC & has two ports i.e. port0 & port1. Out of total 64 pins, 46 pins are available for general purpose input output. Also has an USB port for multiple application.

3.2 Solar Panel It is used to generate the energy. The power from sun rays is absorbed & the generated energy is stored in the 12V battery. In order to achieve maximum power from the sun, we can track sun by using tracking circuit which can be implement from two LDR & DC motor.

217

Cognitive Knowledge Engineering

3.3 Wind Turbine By directing the kinetic energy of wind, we will rotate the wind mill, as wind mill rotates, energy generation will be start at same instance & can save in the same 12V battery.

3.4 Battery Energy generated from solar panel & wind turbine will be used to charge the 12V battery. Also we have to control the flow of energy only in one direction. As energy should not flow again towards the panel & mill.

3.5 Inverter Power supply from battery is in dc form & we have to connect the ac appliances on the battery power & hence we have to convert the power from dc to ac i.e. 12V dc to 230V ac supply. For this purpose we have to employ the step up transformer.

3.6 Zigbee model Zigbee model is used to transmit the information from system to the remote location (Pc). It is a combination of transmitter & receiver. The range of this zigbee model is 70 to 100 meter. Zigbee works on IEEE802.15.4 protocol.

3.7 LCD Display The 16*2 alphanumeric LCD display is used in the system. 16*2 means, 16 columns & 2 rows. In two line 32 characters can be displayed i.e. 16 characters on each line.

4. APPLICATIONS

through comparison of energy usage between the VDPH NLQGV RI KRPH DSSOLDQFHV´ in Proc. IEEEInternational Symposium on Consumer Electronics, Singapore, pp. 1-4,Jun. 2011. [4] Chia-Hung Lien, Hsien-Chung Chen, Ying-Wen Bai, and Ming-%R /LQ³3RZHU PRQLWRULQJ DQG FRQWURO IRU electric home appliances based on power line FRPPXQLFDWLRQ´ in Proc. IEEE International Instrumentation and Measurement Technology Conference, British Columbia, Canada, pp. 21792184, May 2008. [5] Saeed -DKGL DQG /RL /HL /DL ³*ULG LQWHJUDWLRQ RI wind-solar hybrid renewable using AC/DC converters DV '* SRZHU VRXUFHV´ in Proc.World Congress Sustainable Technologies, London, UK, pp. 171177,Nov. 2011. [6] Hayato Yamauchi, Kosuke Uchida, and Tomonobu Senjyu, ³$GYDQFHG 6PDUW +RPH´ in Proc. IEEE International Conference on Harmonics and Quality of Power, Hong Kong, China, pp. 130-135, Jun. 2012. [7] Jinsoo Han, Chang-Sic Choi, Wan-Ki Park, Ilwoo Lee, and Sang-+D .LP ³6PDUW KRPH HQHUJ\ management system including renewable energy EDVHG RQ =LJ%HH DQG 3/&´ in Proc. IEEE International Conference on Consumer Electronics, Las Vegas, USA, pp. 544-545, Jan. 2014. [8] Namsik Ryu, Jae-Ho Jung, and Youngchae Jeong, ³+LJK-efficiency CMOS power amplifier using uneven bias for wirHOHVV /$1 DSSOLFDWLRQ´ ETRI Journal, vol. 34, no. 6, pp. 885-891, Dec. 2012.

This system can be employed in the rural areas for energy generation purpose. Also system can be used to reduce the Energy cost. System will Increase the power generation Capacity of the Nation. Also the system can be used for industrial applications with little modifications. This system will be beneficial to farmers for the irrigation purpose with little modifications.

[9] C. Arm, S. Gyger, J. Masgonty, M. Morgan, J. Nagel, &3LJXHW)5DPSRJQD39ROHW³/RZ-power 32-bit dual_MAC 120 uW/MHz 1.0Vicyflex1 DSP/MCU FRUH´ IEEE Journal of Solid-State Circuits, vol. 44,no. 7, pp. 2055-2064, Jul. 2009.

5. CONCLUSION

[10] +\RXQJVLN 1DP DQG +RRQ -HRQJ ³'DWD VXSSO\ voltage reduction scheme for low-power AMOLED GLVSOD\V´ ETRI Journal, vol. 34, no. 5,pp. 727-733, Oct. 2012

The smart home energy management system can work efficiently in real time conditions. The system is accessible at rural places with number of applications designed for farmers. Statistics of power consumption can be regularly fed to the system using Zigbee protocol. The implementation cost of the system is low. This system developed is friendly to the number of users.

REFERENCES [1] Young-Sung Son and Kyeong-'HRN 0RRQ ³+RPH energy management system based on power line FRPPXQLFDWLRQ´ in Proc. IEEE International Conference on Consumer Electronics, Las Vegas, USA, pp.115-116, Jan. 2010. [2]

Young-Sung Son and Kyeong-'HRN 0RRQ ³+RPH energy management system based on power line FRPPXQLFDWLRQ´ IEEE Trans. Consumer Electron., vol. 56, no. 3, pp. 1380-1386, Aug. 2010

[3] Jinsoo Han, Chang-Sic Choi, Wan-Ki Park, and Ilwoo LHH ³*UHHQ KRPH HQHUJ\ PDQDJHPHQW V\VWHP 218

Biometric: Multimodal System Development

Chapter 45

Review: Unified Approach to Visual Speech and Speaker Recognition Ritesh A. Magre

Ajit S. Ghodke

Dept. of CS and IT Dr. B. A. M. U. Aurangabad

Dept.of MCA, Sinhgad Institute SIBACA Lonavala, University of Pune,Pune, India

ABSTRACT This paper has introduced with Literature Survey, Preprocessing, Different Techniques, Different methods and tools and Classification Techniques. This paper helps in choosing the techniques along with their advantages and disadvantages. 1RZ D GD\¶V VHFXULW\ LV RQH RI WKH LPSRUWDQW issues. This is the way of seFXULW\ IRU LGHQWLI\LQJ VXVSHFW¶V speech on public places such as railway stations, bus stops, and airports. This system also required with acoustic information to overcome the problem of acoustic speech recognition due to noise. Visual Speech combines lip pattern information to identify speech and lip print information to identify speaker. The grooves in the human lips are unique to each person and are used to determine human identification.

Keywords Speech, Visual Speech, Speech Recognition, Lip Reading, Visual Only Speech, Speaker Recognition, Lip Print

1. 1.1

INTRODUCTION Speech

Speech is the way of communication between human beings. Human interact or communicate with each other with the help of speech.

1.2

Visual Speech

Visual speech is defined as a human eye catchable expression of a speaking human face.

1.3

Visual Speech Recognition

The recognition of speech from the visual information only is called as visual speech recognition or lip reading. Video data of a speaker by a camera is recorded. A motion of the speaker face is reduced in large data stream described. The motion of the lips, teeth, chin and tongue are considered [1, 2].

1.4

Speaker Recognition

Speaker recognition is nothing but identifying speaker by machine or recognizing who is speaking. Speaker can be recognized by Audio Visual Information or we can recognize the speaker by visual only information also. Speaker recognition is a system which performs the computing task of validating users claimed identify using features that are extracted from the speech samples. The system automatically detects tracks and identifies a speaker based on visual features extracted from the VSHDNHU¶VPRXWKUHJLRQ

1.5 Unified Approach The meaning of unified approach is nothing but just a unique approach or united approach to achieve a desired goal.

Many methods have been proposed for solving the visual speech recognition problem in the literature despite the variety of existing strategies for visual speech recognition there is still ongoing research in this area to find the most suitable features and classification techniques to discriminate as good as possible between different mouth shapes, but to keep in the same class the mouth shapes corresponding to the same phone produced by different individuals. And require a few processing of the mouth image as possible [2]. The visibility and relations among upper teeth, lower teeth and tongue are very important for human lip reading [3]. As per the research the horizontal and vertical distance between the lips varies for each and every word considering the close proximities of similar sounding word [4]. The lip reading task is even difficult when there is no frontal view of the face. To handle these situations, a pose normalization block is introduced in a standard system and generates virtual frontal views from non-frontal images [5]. The information like phrasing, stress, intonation, emotions, raising and shaping of the eyebrows should also taken into account in visual speech recognition [6]. Lip print characteristics have been widely used in forensics by experts and in criminal police practice for human identification. While examining human lips characteristics the anatomical patterns on the lips are taken into account. The pioneer of Chieloscopy, Professor J. Kasprzak, used 23 lip patterns for finding features of human being. Such patterns (lines, bifurcations, bridges, pentagons, dots, lakes, crossings, triangles etc.) are very similar to fingerprint, iris or palm print patterns. The statistical characteristics features extracted from the lip prints also account for unique identification [12, 13].

2.

PATTERN RECOGNITION

There are different modeling techniques available for speech and speaker recognition [14].The pattern-matching approach (Itakura 1975; Rabiner 1989; Rabiner and Juang 1993) involves two essential steps namely, pattern training and pattern comparison. The essential feature of this approach is that it uses a well formulated mathematical framework and establishes consistent speech pattern representations, for reliable pattern comparison, from a set of labeled training samples via a formal training algorithm. A pattern recognition has been developed over two decade received much attention and applied widely too many practical pattern recognition problem. In the pattern comparison stage of the approach, a direct comparison is made between the unknown speeches (the speech to be recognized) with each possible pattern learned in the training stage in order to determine the identity of the unknown according to the goodness [14].

219

Cognitive Knowledge Engineering

3.

LITRATURE SURVEY Table 2.

Paper Title/ Autor/ Publishing Year/Ref.

Zdenek Krnoul, Milos Zelezny [1]

Mihaela Gordan, Constantine Kotropoulos,Loannis Pitas [2]

Petr Cisar, Milos zelezny, Jan Zelinka, Jana Trojanova September 20007 [3]

Abhay Bagai, Harsh Gandhi, Rahul Goyal, Ms. Maitrei Kohali, Dr. T.V. Prasad April 2009 [4]

Gregory J,Wolff K,Venkatesh Prasad,David G.,Stork and Marcus Hennecke [6]

Advantage / Disadvantage Advantage The Automatic segmentation of Visual Speech Corpora Based. found the time boundaries between consequent phonemes in the corpus Disadvantage The Combination of Visual and Acoustic parameter is not tested. Advantage obtained good word recognition rates as compared to the state of the art results from the literature Disadvantage Advantage The main advantage is the description of both the lip shape and the inner parts of a mouth (upper and lower teeth, a tongue and a gap between teeth).The algorithm of pre-processing of the video data and used methods for both pixel-based and shape based parameterizations were described. Disadvantage Advantage Statistical approach clubbed with neural networks clustering performed significantly better. Beyond the accuracy of their results in reading the lips for letters and words, this neural network model could also be adapted to read sentences by altering the inputs and training the network more intensively Disadvantage the evaluation version does not incorporate all the features required to compute accurate result Advantage VSR in combination with acaustic recognizer reduces the error rate by 75% when compared with the acaustic subsystem alone. Disadvatnage How Visual Speech can be learnt and visual data redundancy can be eliminated are yet not done

Techniques/ Algorithm

Result

Hidden Markov Model, Czeck Audio Visual Speech Corpus Viterbi Algorithm,Baum Velch Algorithm

Author has matched results of the visual segmentation with results of the acoustical on every detected boundary. They specified tolerance thresholds. The thresholds were 20, 50 and 100 ms. Minimal threshold 20 ms is the time distance of two adjacent parameter vectors of the visual signal

Support Vector Machine, Viterbi Algorithm

Word Recognition Rate 90.6%

Shape Based Description, Pixel Based Description, Snakes, Active Shape Model, or Active Appearance Model Fast Fourier Transform, FFT or Discrete Cosine Transform, DCT

Recognition Rate on XM2VTS database is 70.51%. Recognition Rate on UWB-05HSCAVC database is 80.15%

Neural Network,distance clustering approach

With evalution version of software Neurosolution5 the maximum accuracy is 52%

Convolution , Thresholding and edge detection, Classical Algorithm and Learning Algorithm

Video Test Input 50% Correct. Audio Test Input 80% Correct. Audio Video Test Input 95% Correct.

Paper Title/ Autor/ Publishing Year/Ref.

Advantage / Disadvantage

Techniques/ Algorithm

Jonas Beskow

Advantage

KTH text-to-speech

Result A system for spoken man220

Biometric: Multimodal System Development September 1995 [7]

Samir Kumar Bandyopadhyay, S Arunkumar, Saptarshi Bhattacharjee Journal of Current Computer Science and Technology Vol. 2 Issue 1 [2012] 01-08 [8]

Juergen Luettin,Neil A.Thacker, Steve W. Beet [9]

Gajanan Pandurang Khetri, Satish L. Padme, Dinesh Chandra Jain, Dr. H.S. Fadewar, Dr. B.R.Sontakke, Dr. Vrushen P. Pawar 3 November 2012 [10]

Faisal Shafit, Ralph Kricke, Islam Shadaifat, Rolf Rainer Grigat [11]

machine dialogue is currently being developed at the department. The system deals with information on boat traffic, restaurants and accommodation in the Stockholm archipelago

The face can be animated in real time, synchronized with the auditory speech. The facial model is controlled by the same synthesis software as the auditory speech synthesizer. A set of rules that takes co articulation into account has been developed Disadvantage The grouping and parameter dependency of the visemes, and values and dynamic properties of the parameters are all empirically determined and based on relatively limited observations In order to improve the rules, more visual articulatory data is needed.

Advantage Both statistical and anatomical data is considered. Disadvantage -

Advantage ASMs are able to represent lips at various degrees of details and with a small number of parameter. The deformation is purely governed by statistics learned from a training set. Disadvantage The small set of training examples which results in inaccurate estimation of minor modes of variation Advantage FIWAMSR database1 has better performance than IWAMSR database2. Database3 found high recognition rate it main reason behind this database contains nine speech samples collected from different male and female talkers Disadvantage Advantage The accuracy of lip location increased by combining open and closed mouth templates

Gaussian Filter, Sobel Edge Detector, Canny Edge Detector

The proposed method has achieved promising recognition results for well detected lips images

Active Shape Model Gradiant Search, Cost Funtion, Downhill simplex method

The image search was judged by visual inspection a search result was classified as good if the whole lip contour was found within one quarter of the lip thickness deviation. It was classified as adequate if it was found within half the lip thickness deviation and it was classifieds miss otherwise

MFCC and VQ algorithm

database1, database2 and database3 contain 63, 90 and 36 unknown speech samples were tested and yielding an accuracy rate is near about 85.71%, 31.11% and 88.88%

GMM and KNN Classifier

The lip localization accuracy is about 84%.

Disadvantage -

221

Cognitive Knowledge Engineering

4. 4.1

METHODS FOR LIP DETECTION Lip Corner Detection

In this method the positions of both eyes in the face image located. Based on the eye detection lip corners are detected by doing some processing.

4.2

Points Location using AAM

This method gives 75 points on the face and the center of the mouth region is found by determining the centroid of the corresponding mouth points.

4.3 Active Vision Technique for Lip Contour Coding In this technique in each image a contour shape is matched to the boundary of the lips. The space of contours that represent lips is represented by a learned lip contour manifold. During tracking it is tried to find the contour.

4.4

Active Shape Model

This is a technique which represents a lip by a set of labeled point. The point describes the boundary or other significant location of the lip there are two different models are built to characterize lip. The first model describe the outer contour of the lip and second model describes the outer and inner contour of the lip [9]

4.5

Hybrid Parameterization

It is the combination of Pixel Based and Shape Based Parameterization. The Shape Based Parameterization is used to describe the outer lip contour where as Pixel Based Parameterization describes inner part of mouth.

5.

CONCLUSION

In this review, we have discussed different techniques and algorithms developed in each stage of visual speech and speaker recognition system. We also presented the list of algorithms and techniques with their merits and demerits. Through this review it is found that SVM and HMM is used widely for classification and is best among all modeling techniques.

6.

REFFERENCES

[3] Peter Cisar, Milos Zelezny, Jan Zelinka, Jana Trojanova ³'HYHORSHPHQW DQG 7HVWLQJ RI 1HZ &RPELQHG Visual Speech 3DUDPHWHUL]DWLRQ´,6&$$UFKLYH$XJXVW-September 3,2007. [4] Abhay Bagai, Harsh Gandhi, Rahul Goyal, Ms, Maitrei .RKOL 'U 79 3UDVDG ³/LS 5HDGLQJ XVLQJ 1HXUDO 1HWZRUNV´ IJCSNS International Journal ofComputer Science and Network Security VOL.9 No.4 April 2009. [5] Virginia Estellers and Jean-3KLOLSSH 7KLUDQ´ 0XOWL-pose lipreading and audio-visual speech recognition >@ & %UHJOHU DQG 6 2PRKXQGUR ³1RQOLQHDU PDQLIROG OHDUQLQJ IRU YLVXDO VSHHFK UHFRJQLWLRQ´ LQ 3URF ,((( ,&&9 1995, pp. 494-499. [7] Jonas Beskow ³Rule Based Visual Speech Synthesis´ September 1995 [8] Samir Kumar Bandyopadhyay, S Arunkumar, Saptarshi Bhattacharjee´  )HDWXUH ([WUDFWLRQ RI  +XPDQ /LS 3ULQWV´ Journal of Current Computer Science and Technology Vol. 2 Issue 1 [2012] 01-08 >@ MXHUJHQ /XHWWLQ 1HLO $ 7KDFNHU 6WHYH : %HHW ³ $FWLYH 6KDSH 0RGHO IRU 9LVXDO 6SHHFK )HDWXUH ([WUDFWLRQ´ Springer 1996 [10] Gajanan Pandurang Khetri, Satish L. Padme, Dinesh Chandra Jain, Dr. H.S. Fadewar, Dr. B.R.Sontakke, Dr. Vrushen P. Pawar³ Automatic Speech Recognition for Marathi Isolated Words´1RYHPEHU [11] Faisal Shafit, Ralph Kricke, Islam Shadaifat, Rolf Rainer *ULJDW ³5HDO 7LPH /LS 0RWLRQ $QDO\VLV )RU D 3HUVRQ Authentication System Using Near Infrared IllXPLQDWLRQ´ Department of Vision Systems Hamburg University of Technology, Germany [12@ 7VXFKLKDVL @ =GHQHN .UQRXO 0LORV = HOH]Q\ ³7KH $XWRPDWLF Segmentation of the Visual Speech " Department of cybernetics, University of West Bohemia UniverzLWQÕ   14 Plzen, Czech Republic.

[13] - .DVSU]DN - /HF]\QVND %   &KLHORVFRS\ ³+XPDQ ideQWLILFDWLRQ RQ WKH EDVLV RI OLS  3ULQWV´ LQ 3ROLVK  &/. KGP Press,Warsaw, 2001

[2] Mihaela Gordan,Constantine Kotropoulos Ioannis Pitas ³9,68$/ 63((&+ 5(&2*1,7,21 86,1* 6833257 9(&7250$&+,1(6³

[14] Santosh S. Gaikwad, Bharati W. Gawali, Pravin Yannawar ³ $ 5HYLHZ RQ 6SHHFK 5HFRJQLWLRQ 7HFKQLTXH´ International Journal of Computer Applications (0975 ± 8887) Volume 10± No.3, November 2010

222

Biometric: Multimodal System Development

Chapter 46

Review ± Computer Based Lip Reading System for Marathi Language Bhushan S. Kulkarni

Ajit S. Ghodke

Department of Computer Science and IT Deogiri College. Aurangabad (MS), India

Dept.of MCA, Sinhgad Institute SIBACA Lonavala, University of Pune, (MS), India

[email protected]

[email protected]

ABSTRACT This paper is introducing Literature Survey, Different classification techniques such as PCA, HMM, ASM etc. Different methods of feature extraction such as pixel based model, shape based model and combined model. Computer based lip reading system is an attractive research topic and has attracted many researchers in this area of image processing. Many researchers worked on this technique to identify the spoken words only by viewing visual information in speech i.e. lip movements for English and other languages, being Maharashtrians we were inspired to develop a computer based lip reading system for Marathi language that will help people in rural areas of Maharashtra where the people are not able to talk good English. Using this system they can easily interact with computers in the regional language Marathi. This system will also help the deaf and dumb people for communication.

Keywords Lip movements, utterances, lip reading, visual information, Speech Recognition, Image Processing, PCA, Marathi Language, and Lip Reading, HMM.

1. INTRODUCTION Lip reading is a very important technique to interpret the spoken words in speech especially used by the hearing impaired people. They try to understand the speech using lip movement. This technique is also useful when voice is not audible in noisy environments like railway station, bus stand, mall etc. to interpret words by speakers. During speech mouth has different shapes [1].We will use these shapes to identify the words spoken by different people in different conditions. Many researchers have worked on this technique to identify the spoken words only by viewing visual information in speech i.e. lip movements for English and other languages, being Maharashtrian we are inspired to develop a computer based lip reading system for Marathi language that will help people in rural area of Maharashtra where the people are not able to talk good English. Using this system they can easily interact with computers in the regional language Marathi [2]. This system will also help for deaf and dumb people for communication. We will collect sample utterances using video camera from different people (males and females) from various regions of Maharashtra because after some distance Marathi language has variations in phonemes. We will study all the variations for making speaker independent lip reading system. We will extract video frames and acoustic signals from videos. This extracts of video frames are now only visual information in speech. Computer based lip reading system utilizes visual cues related to visual movements during speech, such as tongue , jaw, teeth, and lips movements, to recognize words. The main objective of this system is to accurately decode the visual speech signals into words or into text. The advantage of this system is that visual information is not affected by a lot of noise and does not require sound to recognize .We will use this information for identifying the words spoken by different

Marathi people in Maharashtra. In computer based lip reading system the lip movements are processed to extract features that are later fed into a speech classifiers. The task of speech classifier is to assign the visual features to classes of utterance. The visual features can be categorized into shape based, appearance-based and motion-based features. The shape-based features are parameters extracted from images to describe the shape of mouth, such different horizontal and vertical shapes. 5HVHDUFKHUV XVHG DUWLILFLDO PDUNHUV RQ VSHDNHU¶V IDFH WR HDVH the extraction of lip contours. An artificial marker is less suitable for practical lip-reading applications. Shape-based features can also be extracted through model-based techniques using 2D or 3D model of the lips or mouth.

2. LITRATURE REVIEW For solving the problems in computer based lip reading system many authors proposed many methods such as following: Brgeler uses Time-Delayed neural networks (TDNN) for classification of visual features and the outer lip contour coordinates as visual features. Luettin uses active shape models for representing different mouth shapes, gray level distribution profile (GLDPs) around the lip contours as feature vectors, and finally Hiddden-Markov Models (HMM) for lip reading. Movellan also used HMMs for building visual word models but also used directly the gray levels of mouth images as features preprocessing to exploit vertical symmetry of the mouth [3]. For color image Jamal Ahemad Darggham proposed method in the normalized RGB color scheme. This method called the maximum intensity normalization. The errors in image segmentation reduced more accurately than the pixel intensity normalization. Method for tracking a feature of facial expression, in particular, lip contours proposed by Ying li-Tian by using a multi-state mouth model and combining lip color, shape and lip movements using Gaussian mixture. Principle component analysis (PCA) and mouth changing rate are used for getting the features of the movement of lips. Several preprocessing several steps carried out such as brightness adjustment ,color space ,transforms , face color cutting , facial operation region and lip localization for lip recognition. Ellipse detection method proposed by Yong Hong Xie. Ellipse detected using one dimensional accumulator array. This array accumulates the length information for minor axis of the ellipse. Evaluation of the tangent or curvatures of the edge contours not required in this method [4]. It is the comparison of features extracted from static images with stored templates. According to this method an image, representing the important features of a given utterance, can be captured from the image sequences. Extracted sequence of geometric parameters compared with stored templates of feature sequences. No time wrapping was used to align the sequences. Vector quantization and dynamic time warping (DTW) is extension of this method proposed by Petajan. The quantized images were used in a DTW algorithm to find the best path between the unknown sequence and the stored template. DTW based speech modeling techniques have disadvantage like distance measures between the features. In feature distribution 223

Cognitive Knowledge Engineering and the temporal modeling measures are not considered. This is very specific and is often based on heuristic penalties. *ROGVFKHQH[WHQGHG3HWDMDQ¶VPHWKRGIRUDFRQWLQXRXVVSHHFK reading system using discrete HMM (Hidden Markov Model) . Using a HMM similarity metrics and a clustering algorithm different groups of visemes were detected. The continuous density HMM method used by Nankaku and Tokuda,. The conventional normalization provides a criterion independently of HMM and apply normalization before learning. Maximum likelihood criterion considered in this approach, normalized training is a way of normalization processes for position, size, inclination, mean brightness, and contrast of the lips are integrated with the training of the model [5].

3. CLASSIFICATION TECHNIQUES 3.1 Principle Component analysis Karhunen-Loeve Transform (KLT) method is used for face detection in PCA, it works on dimensionality reduction in face recognition. PCA was exclusively used for face recognition by Turk and Pentland. A set of subspace basis vectors for a database of face images computed by PCA. Images which correspond to a face like structures named Eigen-faces are the basis vectors. Easy comparison of images with the images from the database allowed by projection of images in this compressed subspace. [6]

3.2 Vector quantization

noise. This method is defined as the distribution of apparent velocities in the movement of brightness patterns in an image. 2SWLFDO IORZ LV FRPSXWHG ZLWKRXW H[WUDFWLQJ VSHDNHU¶V OLS contours and location robust visual feature can be obtained for lip movements [12][13].

3.7 Viterbi algorithm This algorithm is dynamic programming algorithm for finding the Viterbi path. Viterbi path is sequence of hidden states that result in a sequence of observed events in context of Markov information sources and HMM [14][15].

4. APPLICATIONS Computer based lip reading system is a significant application in various area as reported in various research in visual speech recognition domain. Some of the area of application is discussed below:

4.1. Lip reading system for hearing impaired people This system helps people who have disability in hearing. They try to understand spoken words by watching lip movements. So with the help of this type of system they can easily communicate to people and can contribute in social activities.

The important aspect of data compression or coding is the process of approximating continuous amplitude signals by digital signals. With respect to distortion or fidelity criterion the number of bits necessary to transmit or store analog data is reduced. Scalar quantization is independent quantization of each signal value while joint quantization is vector quantization or block quantization.[7]

4.2. Identifying speech in noisy area

3.3 Time Delayed Neural Network

4.4 Speech recognition of disordered people

Time delay neural network (TDNN) is architecture of artificial neural network. It is used to work on sequential data. TDNN recognize features independent of time and from larger pattern recognition system. E.g. Converting continuous audio signals into a stream of classified phoneme labels for speech recognition system [8].

Computer based lip reading system has another application to identify spoken words during speech in noisy area such railway station, bus stand, mall etc. where words are not audible and affected by a lot of noise.

4.3 Lip mouse for paralyzed people Piotr Dalka developed a lip mouse to help paralyzed people controlling mouse cursor by using shapes of lip. To recognize words spoken by people who have Dysarthia speech disordered. [16]

4.5 Security Lip reading system used to recognize visual password for computer system [17].

4.6 Enhance Human-Computer Interaction 3.4 Active Shape Models ASM was originally proposed by Cootes. The modeling technique is similar in spirit to Active Contour Models, or snakes, proposed by Kass but the advantage of ASM is that instances of models can only deform in ways found in a training set. That is, ASM allows considerable variability but are still specific to the class of objects or structures they intend to represent [9] [10].

3.5 Hidden Markov model From last 30 years HMMs is used in speech recognition. It is a statistical tool. It is also used in recognition of biological sequence analysis [11]. They are suitable to handle discrete sequences of varying sizes. Efficient training algorithm and efficient decoding algorithm is available. Which provides optimal sequence of states associated with a given sequences of low level data.

3.6 Optical ±flow analysis Optical flow analysis is new multi modal speech recognition method, evaluates its robustness to acoustic and visual speech

The main goal of HCI application is to make working easy with a computer, intuitive and effective as possible [18].

5. METHODS Very first step in lip reading system is feature definition and extraction. All extracted features describe the shape of lips, surroundings region properly. The feature extraction method can be grouped as follows x Pixel Based Methods (Appearance Models or Data Based Models). x Pixel Based Transformation methods. x Model Based transformation [19].

5.1 Pixel based methods In this method Pixels values are considered as features. Image is reduced in ROI (Region of Interest). This ROI is rectangle of size N * M or ellipse or other shape. These methods have weak resistance against noise, condition change and affine transformation of their high dimension they are very GHPDQGLQJ RI FRPSXWDWLRQDO WLPH 7KHUHIRUH D 3L[HO¶V transformations of ROI are applied. ROI is reduced to a rectangle on lips or around lips mostly, sometimes around the

224

Biometric: Multimodal System Development small parts of face like chin or nose. Rarely the whole face is used [19].

5.2 Pixels transformation based methods Low pass filter is used for sub sampling and image subtraction. For dimensionality reduction image sieve is used. PCA (Principal Component Analysis),Linear Discriminant Analysis(LDA), Discrete Cosine, Wavelet, Curvelet Transformation and many traditional image transformation techniques are used it is also called techniques for image compression. For working with different processing stages PCA is used by many people. Form other transformation cosine and Hadmard are used. Harr and curvelets in opposite to the above presented transformations stand geometric type of features: 1. Vertical and Horizontal distance in lips. 2. Perimeter of lip contour. 3. Inner mouth area. 4. Visual tongue and teeth area. These type of feature used in previous work [19].

[3]

[4]

[5]

[6]

[7]

5.3 Model based methods There is special category of features that are parametrical and statistical models. Many of these used as preprocessing steps for searching for face and lips ROI initialization. Snakes and Active Shape Model is well known. Many approaches are combined to get robust features. The combinations like geometric and PCA. In PCA with ASM are used. This mutual combination is called Active Appearance Model (AAM). While designing the special features following things must be considered. 1. For each frame in video vectors must be extracted. 2. ROI is set more accurately. 3. Silence elimination. 4. System robustness and generalization [19].

6. CONCLUSION This paper brings the idea of how speech can be recognized using several transforms and many classifications methods which are required to come up with desired output. In this paper we study about different classification methods proposed by many people for feature classification in lip reading. For feature extraction and definition of lip during speech many methods were proposed like pixel based methods, model based transformation, combined methods. Using these classification techniques and feature extraction methods computer based lip reading system can be developed for English and other language, so taking ideas from these methods we plan to develop the computer based lip reading system for Marathi language. This system is useful for Marathi people in different areas such as Human computer interaction, security etc. This system also brings a new way of communication for rural people in Maharashtra. This system further can be made as speaker independent, robust and for continuous speech recognition.

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

7. REFERENCES [1] 6ODPD 3DWKDQ $UFKDQD *KRWNDU³ Recognition of spoken English phrases using visual IHDWXUHVH[WUDFWLRQDQGFODVVLILFDWLRQ´'HSDUWPHQWRI Computer Engineering,Pune Institute of Computer Technology,Pune,India,ISSN NO:0975-9646. [2] Misba Hudewale, Sukhbir Gill, Sumit Ahirwar ³,VRODWHG:RUG5HFRJQLWLRQ,Q0DUDWKL´'HSDUWPHQW 2I (OHFWURQLFV $QG 7HOHFRPPXQLFDWLRQ-630¶V

[16]

[17]

RSCOE. Savitribai Phule University,Pune,Maharashtra,IJETAE Volume 5,1 january 2015. Mihaela Gordan,Constantine Kotropoulos,Ioannis 3LWDV³9,68$/63((&+5(&2*1,7,2186,1* 68332579(&7250$&+,1(6´ Kamil S. TALHA, Khairunizam WAN, Viratt &KLWWDZDG 6.=D¶ED DQG 0 1DVLU $\RE Zuradzman M. Razlan, and Shahriman AB ³([WUDFWLQJ )HDWXUHV 3RLQW of Lip Movement for Computer-EDVHG /LS 5HDGLQJ 6\VWHP´ ,-00(IJENS Vol:14 No:02 Salah Werda, Walid Mahdi and Abdelmajid Ben +DPDGRX ³/LS /RFDOL]DWLRQ DQG 9LVHPH &ODVVLILFDWLRQIRU9LVXDO6SHHFK5HFRJQLWLRQ´,-&,6 Volume 5 No 1 April 2007 +DUGLN .DGL\D ³&omparitive Study on Face Recognition Using HGPP, PCA, LDA, ICA and 690´ *OREDO -RXUQDO RI &RPSXWHU 6FLHQFH DQG Technology Graphics & Vision Volume 12 Issue 15 Version 1.0 Year 2012 Jhon Makhoul, Salim Roucos Felllwo IEEE ³9(&725 48$17,=$7,21 ,1 63((&+ CODI1*´ 0018-9219/85/1100-1551901.00 81985 IEEE. 'U%63UDGHHS1DYHHQ/³:HDWKHU3UHGLFWLRQE\ Mass Neural Networks through an Integrated Model DSSURDFK´ ,QWHUQDWLRQDO 5 ' GHSW /LQ\L 7RS Network Co. Ltd 2Dept. of ISE, Rajarajeswari College of Engineering, ISSN-ONLINE 2455-1457. Cootes T, Taylor C, Cooper D, Graham J. Active Shape Models - Their Training and Application. Computer Vision and Image Understanding, January 1995, Vol. 61, No. 1, pp. 38-59 *KDVVDQ +DPDUQHK ³$FWLYH 6KDSH 0RGHOV Modeling Shape Variations and Gray Level Information and an Application to Image Search and &ODVVLILFDWLRQ´ 6DP\ %HQJLR ³$Q $V\QFKURQRXV +LGGHQ 0DUNRY Model for Audio-9LVXDO6SHHFK5HFRJQLWLRQ´'DOOH Molle Institute for Perceptual Artificial Intelligence (IDIAP) CP 592, rue du Simplon 4, 1920 Martigny, Switzerland. Berthold K.P. Horn and Brian G. Rhunck, ³'HWHUPLQLQJ 2SWLFDO )ORZ´ $UWLILFLDO ,QWHOOLJHQFH Laboratory, Massachusetts Institute of Technology, Cam bridge, MA 02139, U.S.A. [17] Tomoaki Yoshinaga, Satoshi Tamura, Koji Iwano, DQG 6DGDRNL )XUXL ³$XGLR-Visual Speech Recognition Using Lip Movement Extracted from Side-)DFH ,PDJHV´ Department of Computer Science, Tokyo Institute of Technology. 0 6 5\DQ DQG * 5 1XGG ³7KH 9LWHUEL $OJRULWKP´ 'HSDUWPHQW RI &RPSXWHU Science, University of Warwick, Coventry, CV4 7AL, England, 12th February 1993. . 6 $UXQODO DQG 'U 6 $ +DULSUDVDG ³$1 ()),&,(17 9,7(5%, '(&2'(5´ &HQWHU IRU Cognitive Technologies, Bangalore, Karnataka, India. International Journal of Computer Science, Engineering and Applications (IJCSEA) Vol.2, No.1, February 2012 Elham S. Salam,Reda A EI-khoribi,Mahmoud E. 6KRPDQ ³$XGLR-Visual Speech Recognition for 3HRSOH ZLWK 6SHHFK 'LVRUGHUV´ ,-&$ 8887,volume-96-No.2, June 2014. $KPDG %$+DVVDQDQW³9LVXDO Password Using $XWRPDWLF /LS 5HDGLQJ´ ,7 GHSDUWPHQW 0X¶WDKKarak, Jordan,61710. 225

Cognitive Knowledge Engineering [18] 3LRUWU 'DOND $PGUH]M ³+80$1-COMPIUTER BASED VISUAL LIP MOVEMNET AND *(6785( 5(&2*1,7,21´ ,-&$9RO 1R SS 124-139,2010.

[19] 6WHIDQ %DGXUD 0LFKDO 0RNU\V ³)HDWXUH H[WUDFWLRQ for automatic lips reading system for isolated vowels´ The 4th International Virtual Scientific Conference on Informatics and Management Sciences March, 23. - 27. 2015.

226

Biometric: Multimodal System Development

Chapter 47

DATA RECOVER FROM SECONDARY STORAGE DEVICE AFTER MULTIPLE TIMES DELETING FORMATING AND WIPING Manisha M. Phopse

Charansing N. Kayte

Department of Forensic Science , Yeshwantrao Chavan Science Institute, Satara Dist. Satara.

Dept. of Digital and Cyber Forensic, Government Institute Of Forensic Science Aurangabad

[email protected]

[email protected]

ABSTRACT This paper we are show how Data recovery is the process of salvaging data from damaged, failed, corrupted, or in-accessible Secondary storage media when it cannot be accessed normally. Such cases can often be mitigated by disk partitioning and consistently storing valuable data files (or copies of them) on a different partition from the replaceable OS system file [1]. The main objective of this research is to data recovery from multiple overwritten [2], deleted [1] and formatted secondary storage media. In our research we have done data recovery from secondary storage media using forensic tools in different ± different situations.

Keywords:

RAM, Big Data, CD/DVD, UFD, USB, FAT,

All processed of recovering partition which is finding become disappeared & become unallocated space on disk. This is one type of partition error in data recovery process.

1.1.1.2 Formatted Recovery(FR) Formatted Recovery process of recover data from hidden index but not overwritten the exiting file. There is generally two type of disk formatting low level formatting- is the process of recording mark track and sector on drive. High level formattingis the process generating new file system or emptying file system on drive.

1.1.1.3 Damaged Partition Recovery(DPR)

PAN-DRIVE

DPR processed of recovering partition which asked about formatting unrecognized file format.

General Terms:

1.1.1.4 Deleted Partition Recovery(DPR)

This paper we are introduce how data recover from secondary storage device by multiple time erase data from hard disk and other storage device.

DPR deleted partition means space on hard disk which is unallocated and new partition created.

1. INTRODUCTION Data recovery means retrieve deleted or corrupted data form Secondary storage media. It is required when storage media physical damage, logical damage in any situations. Data recovery often faces challenges with various kinds of storage devices it can and does address even the most typical data loss circumstances. The data which appear to be lost and inaccessible from computer about some miss-happing such as hard disk failure, corrupt files, other data loss issues, system crash and some other problem there is need to protect data or information from such type of damage in that situation data recovery[2].

2. DATA STORING MECHANISA ON EXTERNAL STORAGE DEVICE 2.1 Data storing mechanism on hard disk Data is stored on the hard disk is magnetically the form of 0 and 1 binary code. The part of the hard disk that stores the data is known as platter, which is divided into billions of parts called tracks. Platter can be one or more in number depending on the size of hard disk and its storage capacity. Data is stored digitally in the form of tiny magnetized regions on the platter where each region represents a bit [3].

1.1 Media Type Currently popular technologies include hard drives, CD/DVD drives, flash-memory cards, and USB flash drives. Other names for a USB flash drive (UFD) include USB key, pen drive, ThumbDrive®, DiskOnKey®, and JumpDrive®[3].

1.1.1

Type of Data Recovery

1.1.1.1 Lost Partition Recovery (LPR) Figure 1 Internal Structure of Drive. 227

Cognitive Knowledge Engineering

2.2 Mechanism of hard disk read the data The operating system works out where the data is on the disk. It first reads the FAT (File Allocation Table) at the beginning of the partition. This tells the operating system in which sector on which track to find the data. With this information, the head can read the requested data, Heads are energy converters they transform electrical signals to magnetic signals, and magnetic signals back to electrical ones again. Applying a magnetic field to a coil will cause an electrical current to flow; this is used when reading back the previously written information. Set of data to be read is sequentially located on the disk [3].

a

a

b Figure 3 Hard Disk Data Writing Mechanism (a, b). b

2.4 Data storing mechanism on Flash drive (pen drive)

Figure 2 Hard Disk Data Reading Mechanism (a, b) [5].

The NAND flash memory chip is the core of the flash drive, PHPRU\ LV DQ REMHFW WKDW DOORZV XV WR ³VHOHFWLYHO\ VWRUH RU VHOHFWLYHO\UHWULHYHELWVRILQIRUPDWLRQ´7KHIODVKGULYHEHORQJV to Transistor-EDVHG VHPLFRQGXFWRU PHPRULHV µ7UDQVLVWRUV¶ however operate electrically. The flash drive therefore has no moving parts and information stored on the NAND flash memory. The NAND flash memory derives that memory cells are arranged are connected in series. When we save a file on the flash drive function of the two transistors known as the ³&RQWUROJDWH´DQGWKH³)ORDWLQJJDWH´>@

2.3 Mechanism of hard disk writes the data Magnetic field is placed on the tiny field in one of these two polarities: N-S. In the one direction (like N-S) can represent the µ¶ZKLOHWKHRSSRVLWHRULHQWDWLRQ6-1 UHSUHVHQWV³´6WRULQJ the data on platter they are called tracks and which is made up of sectors. The Actuator looks through a table of stored data locations to find the where the chosen data is located. File Allocation Table (FAT) is a map of information stored in hard disk, it tells which part of hard disk is free and which is still free. The correct sectors pass beneath the head, the magnetic fields from the bits induce resistively changes in the sensitive materials located in the reading elements within the head. As the head passes over the surface of the disk, the material changes resistance as the magnetic fields changes corresponding to the stored patters on the disk [5].

a

228

Biometric: Multimodal System Development

3

10

100%

100%

100%

100%

100%

4

20

100%

100%

100%

100%

100%

5

30

100%

100%

100%

100%

100%

6

40

100%

100%

100%

100%

100%

7

50

100%

100%

100%

100%

100%

8

80

100%

100%

100%

100%

100%

9

100

100%

100%

100%

100%

100%

b Figure 4-Inside Flash Drive (a, b).

2.5 Saving data on the flash drive [writing] When file save on flash drive, a voltage is applied to the control gate which then sends electrons from the source to start flowing towards the drain. In the process of the flowing, the electrons gain energy to penetrate the oxide layer and gets stored in the floating gate. the floating gate forms a negative charge (that is, data is recorded on the gate) the floating gate is coated with non-conductive material This means that even when we plug out the flash drive, the data is safely stored on the floating gate. Floating gate is coated with non-conductive memory; due to this data will not be lost even when there is no power [5].

Table 2 Formatted pen drive recovery % of recovery Stellar Icare power Phonix

Laze Soft

100%

100%

100%

100%

100%

5

100%

100%

100%

100%

100%

3

10

100%

100%

100%

100%

100%

4

20

100%

100%

100%

100%

100%

5

30

80%

80%

70%

65%

70%

6

40

70%

75%

60%

60%

60%

7

50

60%

60%

55%

50%

50%

8

80

30%

30%

20%

20%

30%

9

100

File corrupt

File corrupt

10%

0%

0%

Sr.

No.of Times deleted

1

1

2

Eases us

Figure 5- Data Storing and Writing Mechanism in Pen Drive.

2.6 Methodologies and Algorithm In our research we have taken three different pen drive of same memory size mainly two 4GB and 8GB of scandisk. We have fill all three pen drives with same data at a time and after that delete all data from first pen drive , second pen drive format using DOS command and third pen drive wiping using ccleaner. Then we select best forensic tools for data recovery from all three Pen Drives. When we recover data from both pen drives we compare data of all three pen drives. Same situation follow number of times we mention in following table. Table 1 Deleted pen drive recovery. No. of Times Deleted

Eases us

% of recovery Stellar Icare power Phonix

Laze Soft

1

1

100%

100%

100%

100%

100%

2

5

100%

100%

100%

100%

100%

Sr. No

Table 3 ± Wiping Pen Drive Recovery Sr.

No.of Times deleted

Eases us

% of recovery Stellar Icare powe Phonix r

Laze Soft

229

Cognitive Knowledge Engineering

1

Gutmann 35 passes

2

the data can be recovered from deleted and formatted pen drive is depends on the size of the data and number of times the data was overwritten. We found result in our research more chance to recovered data from deleted pen drive as compare to formatted drive and wiping.

0%

0%

0%

0%

0%

Random write

0%

0%

0%

0%

0%

3

DOD 3 passes

30%

30%

0%

0%

0%

There is future scope and one biggest challenge to develop a technique which recover data from entirely wiped drive and make data recovery possible from such condition.

4

DOD 7 passes

0%

0%

0%

0%

0%

5. REFERENCES

3. RESULT We have use various IRUHQVLF GDWD UHFRYHU\ VRIWZDUH¶V OLNH Eases data recovery, stellar phoenix, icare data recovery, power data recovery and laze soft. When we have recovered data from all pen drives and compare data of each pen drives. We recovered 100% data from deleted and formatted pen drive but from third (wiping) pen drive only 30% data recovered. Same situation follow with all three pen drives number of times. We have found different ± different results.

[1] 3HWHU*XWPDQQ³6HFXUH Deletion of Data from Magnetic and Solid-6WDWH0HPRU\´6L[WK86(1,; Security Symposium Proceedings, San Jose, California, July 22-25, 1996. [2] Khurshudov, A. The Essential Guide to Computer Data Storage. USA: Prentice- Hall PT. 2001. Print. [3] Thomas M. Coughlin, Digital Storage in Consumer Electronics, chapter 2, pp30, [4] Elsevier .Jan Axelson , USB mass storage,chaper 1,pp9 [5] Jiles, David: introduction to magnetism and magnetic material, 2nd edn. Chapman & Hall, Boca Raton (1998)

4. CONCLUSION AND FUTURE WORK We conclude that we worked on different - GLIIHUHQWVRIWZDUH¶V for data recovery in different situation. We have observed that

230

Biometric: Multimodal System Development

Chapter 48

,PDJH,QFRQVLVWHQF\'HWHFWLRQ%DVHGRQ*UDGLHQW 'LUHFWLRQ Mahale Vivek Hilal Dr. Babasaheb Ambedkar Marathwada University Aurangabad, (M.S), India

[email protected]

Pravin Yannawar

Ashok T. Gaikwad

Department of CSIT Dr. Babasaheb Ambedkar Marathwada University Aurangabad, (M.S), India [email protected]

Institute of Management Studies and Information Technology, Aurangabad, (M.S), India [email protected]

ABSTRACT The Digital Image inconsistency in light source direction for discover forgery in image are active research work to build the identification and distinguish between original and forge image. This paper focus on detect the inconsistency of digital image by many steps. Steps start with input image convert into grayscale then divided to overlapping blocks and calculate gradient and Standard Deviation for each block. The result of this work tested on CoMoFoD Image Database and discover good results.

2. PROPOSE OF STUDY TO DETECT THE INCONSISTENCY IN IMAGE The Inconsistency nowadays is the field of many research to detect the copy and paste in the image, the steps including in our system as show in the fig 1.

General Terms Image processing, Forgery, Image Analysis

Keywords Forgery Detect, Gradient , Standard Deviation

1. INTRODUCTION Recent , Digital image are easily misrepresent with the help of image processing software. In systematic forensic sciences, all manner of physical evidence is analyzed .It is real hard for humans to discover whether the digital image is original or misrepresent. There is rapidly increase in digitally misrepresented forgeries in mainstream media and on internet. Today from magazines to fashion industry and media, courtroom, photo prankster, political campaigns misrepresent photographs are increasing in past years [1]. One of the most common type of tempering in digital image is copy move . in copy move tempering one part of the image is copied and past to another part of the same image. Human eye cannot discover suspect artifact easily because copied part is come from same image so image properties will be compatible with rest of the image[2] . forgery detection methods can be categorized as active and passive[4].active method is concerned with data hiding where some code is embedded in image. It has two types watermarking and digital signatures [5][6]. Passive method no need prior information just the image itself. Passive method base on forgery type dependent and forgery type independent. Forgery type dependent has again two type copy- move detection and image splicing detection. Forgery type independent has also two type retouching detection and lighting conditions[7]. The paper organized into four section , in section 2 the propose of study are present. In section 3 the excremental and result are given. The conclusion and future work are given in section 4.

Fig 1:The propose of study The propose of studies show in which including two phases. Propose studies starting with first phase which is training phase, which include input image , which I took from CoMoFoD Image Database for Copy Move Forgery Detection [8]. In this dataset 200 images is present. After taking input png image convert it into gray scale image with the help of equation 1 ‫ ܫ‬ൌ Ͳʹͻͺͻ ൈ ܴ ൅ ͲǤͷͺ͹Ͳ ൈ ‫ ܩ‬൅ ͲǤͳͳͶͲ ൈ ‫ ܤ‬-----Æ(1) Then, segment that gray scale image in 2 x 2 , after that calculate the gradient for intensity direction[9] [10].Gradient µP¶LV ݉ ൌ

ο‫ݕ‬ ‫ݕ‬ଶ െ ‫ݕ‬ଵ ൌ   െ െെ՜ ሺʹሻ ‫ݔ‬ଶ െ ‫ݔ‬ଵ ο‫ݔ‬ 231

Cognitive Knowledge Engineering Then calculate Standard Deviation of gradient direction to detect which block has inconsistency in light of direction. σ࢔ ഥሻ૛ ࢏స૚ሺ࢞࢏ ି࢞

ࡿ࢞ ൌ  ට

࢔ି૚

-----------Æ(3)

In formula of Standard Deviation n = the number of data points , ‫ݔ‬ҧ ൌ ݄ܶ݁݉݁ܽ݊‫ݔ݂݋‬௜ and xi= each of the value of the data In the testing phase it is same steps in training phase with addition comparison steps between training set and testing sets the comparison is done by subtraction between Standard Deviation value for both cases. if the result of subtraction EHWZHHQEORFNV LV]HURWKDW¶VPHDQVEORFNLVFRQVLVWHQFH otherwise it is inconsistence.

Third step which is divided gray scale image into 2 x 2 blocks and calculate the gradient for each block by direction of intensity value separately then the Standard Deviation are calculated for each blocks for extract statistical features which help us to detect the consistency and inconsistency by subtract Standard Deviation for each block for training and testing images. The figure 4, show the example of the result of divided the image into 2 x 2 block and gradient direction of each block.

3. Experimental and Result In this work the experimental tested on CoMoFoD Image Database for Copy Move Forgery Detection [4]. In this dataset 200 images is present image size is 512 x 512 pixel. Image is color type. we select 20 image for tested our experiment . Then after fig.2 shows sample from our database which we used.

(a)

(b)

(c)

(d)

(a)

Fig 4: divided image 2x2 blocks with gradient (a) image 1 original (b) image 1 forge (c) image 2 original (d) image 2 forge

(b)

Fig 2: sample image from database (a) original (b) forge The second step of our process is to convert input and testing image to gray scale by using equation 1.The result of this steps show in figure 3. Which read the image from database and convert it to grayscale image for use it the next steps.

The Standard Deviation values for original image and forge image for each block shown in the table 1

Table 1 : The Standard Deviation values for first sample Sample Image

Original - 1

Forge - 1

(a)

(b)

Fig 3: grayscale image (a) original (b) forge

block

std

1

102.9916

2 3 4

103.1731 103.0468 102.0575

1

102.9916

2 3 4

103.2752 103.0468 102.0575

The table above show Standard Deviation of inconsistency and consistency image for first sample which we selected for test our experiments , the image 1 is original and also we take the inconsistency image of original. The result of our experiment shows which blocks has consistency and inconsistency by taken the differences between Standard Deviation for each block. The 232

Biometric: Multimodal System Development fig 5 shows the result of subtracted the Standard Deviation values form each block of original and forge, from the graph below we see that the second block has inconsistency ,means that the block 2 has manipulated.

4. CONCLUSION The study of image inconsistency to discover the manipulated in the image is active research and more security to detect the difference between original and forge image. This study we implement image forgery detection algorithm on CoMoFoD Image Database. The main steps for our algorithm which convert input image into grayscale image and divided into overlapping blocks then calculate the gradient and Standard Deviation for each block in train and testing sets. Then we compare each block separately to discover which block has change in values (means pixel by pixel compare). The result shows the block which has inconsistency for our image test. The future work may extend to use textual feature to show discover inconsistency from the digital image which improve.

5. ACKNOWLEDGMENTS Fig 5: graph of Standard Deviation for each block of first sample image Table 2 : The Standard Deviation values for second sample Sample Image

Block

Std

1

104.0817

Original - 2

2 3 4

103.2722 103.4056 103.9684

1

104.0817

Forge - 2

2 3 4

103.2722 103.4056 103.9826

For another case which we selected the second image from our database, the Standard Deviation values of inconsistency and consistency image for second sample are used for test our experiment. The result show which block has consistency and inconsistency as show in fig 6 . We see that the block 4 has inconsistency ,means that manipulated .

Our thanks to Video communication Library to provided us database (CoMoFoD Image Database) which we test our experiments on it .

6. REFERENCES [1] Hany Farid, 2009, ³,PDJH )RUJHU\ 'HWHFWLRQ´ ,((( SIGNAL PROCESSING MAGAZINE, pp. 16-25. [2] D. Soukal and j. Lukas, 2003, ³'HWHFWLRQRIFRS\± move IRUJHU\LQGLJLWDOLPDJHV´3URFRf DFRWS , Cleveland. [3] Jen ±Chun Lee, Chien ± Ping , Wei ± Kuei Chen, 2015 ³'HWHFWRIFRS\-move forgery using histogram of oriented JUDGLHQWV´ (/6(9,(5 ,QIRUPDWLRn Science , vol.321, page 250-262. [4] Richard Dosselmann and Xue Dong Yang, 2009´ Determining Light Direction in Spheres using Average Gradient´ echnical Report TR-CS 2009-1 ,ISBN 978-07731-666-6 (on-line) ISSN 0828-3494. [5]

S. Katzenbeisser and F. A. P. Petitcols, 2000, ³,QIRUPDWLRQ 7HFKQLTXHV )RU 6WHQRJUDSK\ $QG 'LJLWDO :DWHUPDUNLQJ´1RUZRRG.

[6] Z. Zhang, Y. Ren, X. J. Ping, Z. Y. He and S. Z. Zhang, 2008,³$ VXUYH\ RQ SDVVLYH-blind image forgery by doctor PHWKRG GHWHFWLRQ´ 3URF 6HYHQWK ,QW &RQI RQ 0DFKine Learning and Cybernetics,pp. 3463±3467. [7] Saba Mushtaq and Ajaz Hussain Mir ,2014,³Digital Image Forgeries and Passive Image Authentication Techniques: A Survey´ International Journal of Advanced Science and Technology , Vol.73 , pp.15-32 [8] http://www.vcl.fer.hr/comofod/comofod.html [9] Z. Zhang, Y. Ren, X. J. Ping, Z. Y. He and S. Z. Zhang, 2008,³$ VXUYH\ RQ SDVVLYH-blind image forgery by doctor PHWKRG GHWHFWLRQ´ 3URF 6HYHQWK ,QW &RQI RQ 0DFKLQH Learning and Cybernetics, pp. 3463±3467. [10] 0LFDK.-RKQVRQ+DQ\)DULG³([SRVLQJ'LJLWDO )RUJHULHVE\'HWHFWLQJ,QFRQVLVWHQFLHVLQOLJKW´$&0 Multimedia and Security Workshop,05 New York , USA, ACM 1-59593-032-9/05/0008.

Fig 6: graph of Standard Deviation for each block of second sample image

233

Cognitive Knowledge Engineering

Chapter 49

Feature Level Fusion for Fingerprint using Neural Network for Person Identification Siddiqui Almas

Lothe Savita A.

Telgad Rupali L.

Deshmukh P. D.

Dr. B.A.M. University, Aurangabad.

Dr.B.A.M. University, Aurangabad

Dr. B.A.M. University, Aurangabad.

siddiqui.almas29@g mail.com

[email protected] m

rupalitelgad@rediffm ail.com

0*0¶V'U* 125

High

Figure-5: Membership Function for PPG and Random Glucose

Post Prandial Glucose (PPG) measures the glucose level after 2 hours of meal. The two tests are most commonly used for diagnosis. But in many cases random sugar test, also required for confirmation. The range of values of PPG and random sugar are shown in Table-2. Table-2: Range of PPG and Random Sugar Values (mg/dl)

Specifications

< 140

Normal

140 to 199

Medium

>= 200

High

The criteria of glucose levels for diagnosis of type-2 diabetes are: [6] FBG >= 126 mg/dl OR PPG >= 200 mg/dl OR Random glucose >= 200 mg/dl In Prediabetes, sugar levels in the blood are above the normal range, which is not sufficient to classify as a diabetes.[4] The measures of blood glucose for prediabetes are: FBG between 100 and 125 mg/dl OR PPG between 140 and 199 mg/dl

Figure-6: Membership function for DMRisk The layer-2 is fuzzification layer where the input values are fuzzified. In third layer, the rules are formulated as per WHO guidelines. The rules for prediabetes and type-2 diabetes are given below. IF FBG if medium and PPG or Random glucose is medium then DMRisk is Prediabetes If FBG is high and PPG or Random glucose is high then DMRisk is Type-2 diabetes 296

Biometric: Multimodal System Development The outputs of these rules are combined in layer-4 using fuzzy union operator and finally it is defuzzified in layer-5.

4. EXPERIMENTAL RESULTS To test the performance of this proposed system the data of 350 patients collected for blood sugar values and implemented in fuzzy logic toolbox of MatLab. The Sugeno type FIS is used for implementation. The output membership function in this type is either linear or constant. 193 records used for training purpose and remaining 157 for testing. Out of which 50 are non-diabetic, 55 are prediabetic and 52 are of type-2 diabetes. Table-3 shows the class-wise diagnosis of the patients. Table-3: Classification Percentage of the System

[3] R. David Leslie, M. Cecilia Lansang, Simon Coppack, /DXUHQFH .HQQHG\ ʊ'LDEHWHV ± Clinical Desk 5HIHUHQFH‫  ۅ‬0DQVRQ 3XEOLVKLQJ /WG &5& 3UHVV ISBN: 978-1-84076-158-0. [4] R.P. Ambilwade, R.R. Manza, Bharatratna P.Gaikwad, µ0HGLFDO ([SHUW 6\VWHPV IRU 'LDEHWHV 'LDJQRVLV $ 6XUYH\¶,QWHUQDWLRQDO-RXUQDORI$GYDQFHG5HVHDUFK in &RPSXWHU 6FLHQFH DQG 6RIWZDUH (QJLQHHULQJ¶ ISSN:2277 128X, Vol. 4 Issue 11, Nov. 2014. [5] International Diabetes Federation 2016, http://www. diabetesatlas.org/ IDF_Atlas2015_UK.pdf

Class

No. of Patients

Diagnosed by the System

Percentage

[6] World Health Organization 2011. http://www.who.int/ mediacentre/factsheets/fs312/en/2011

Non-Diabetic

50

41

82%

Pre-diabetes

55

43

78.18%

[7] Standards of Medical Care in Diabetes, 2016, American Diabetes Association

Type-2 Diabetes

52

44

84.61%

[8] 0LFKDHO1HJQHYLWVN\³$UWLILFLDO,QWHOOLJHQFH± A Guide to ,QWHOOLJHQW6\VWHPV´3HDUVRQ(GXFDWLRQ,QF

The overall classification percentage is 81.59%, which is good to predict the diabetes from only blood sugar levels.

5. CONCLUSION The system proposed in this paper is novel neuro-fuzzy system used for diagnosis of type-2 diabetes and prediabetes. Diabetes is a silent disease which recognize mostly after age of 40 years. The symptoms of diabetes are not so obvious, but develop slowly during years. It is due to the fact that the body cannot produce insulin or does not use it properly. Due to which the amount of glucose in blood rises. The diagnosis criteria for diabetes are mainly based on blood sugar levels. This diagnosis process propose in this paper is based on the fuzzy if-then rules and neural network training algorithm. The layered approach of neural network and fuzzy inference process is combined to get the benefits of both the technology. The system is implemented in Suegno type FIS and tested on 157 patients records. The classification accuracy obtained is 81.59% and is relatively good. The accuracy can be further increased by adding other inputs to the system like symptoms, past and present history of patients and risk factors. There is further scope to improve the performance of the system using various ANN model and FIS.

6. REFERENCES [1]

DQG $SSOLFDWLRQV¶, New Delhi, Prentice-Hall of India Private Ltd.

Anupam Shukla, Ritu Tiwari, Rahul Kala, 2010, ͚ZĞĂů>ŝĨĞŽĨ^ŽĨƚŽŵƉƵƚŝŶŐ͕͛ZWƌĞƐƐ͕dĂLJůŽƌ and Francis Group, LLC, Auerbach Publications.

[9] 3DGK\ 13  µArtificial Intelligence and Intelligent Systems¶, New Delhi, Oxford University Press [10] .OLU *HRUJH - )ROJHU 7LQD $  µFuzzy Sets, Uncertainty and Information¶, New Delhi, Prentice-Hall of India Private Ltd. [11] 3DQRV6DUDILGLVµ([SORULQJWKH1HXUR-Fuzzy from Theory WR3UDFWLFH¶Research Article [12] * 6KDUDGD 'U2% 9 5DPDQDLDK  µ$Q $UWLILFLDO Intelligence Based Neuro-Fuzzy System with Emotional ,QWHOOLJHQFH¶ ,QWHUQDWLRQDO -RXUQDO RI &RPSXWHU Application, Vol 1- No. 13, ISSN: 0975-8887 [13] Sujit Das, Pijush Kanti Ghosh, Samarjit Kar, µ+ypertension diagnosis a comparative study using fuzzy expert system and neuro fuzzy system¶  IEEE International Conference on Fuzzy Systems, 7-10 July 2013, ISSN :1098-7584 [14] www.web.cecs.pdx.edu/~mperkows/CLASS_479/2013/lect ures/2012-1161.Neuro-Fuzzy Systems.ppt [15] http://ftp.it.murdoch.edu.au/units/ICT219/Lectures/03B219 Lect_Week11.pdf R.P. Ambilwade, R.R. Manza, Ravinder Kaur, µDevelopment and Design of the Input Parameters of Expert System for Diagnosis of Diabetes¶  1&(7&6$2013, National Conference on Emerging Trends in Computer Science and Applications, Fergusson College, Pune, India, ISBN- 978-93-5137-872-3, Dec. 7,8 2013

[2] 5DMDVHNDUDQ 6 3DL 9LMD\ODNDVKPL *$  µNeural Networks, Fuzzy Logic and Genetic Alogrithms Synthesis

297

Cognitive Knowledge Engineering

Chapter 62

Comparison of Enhancement Techniques for the Betterment of Dental Radiograph Nirupama S. Patodkar

Rita B. Patil

Prapti D. Deshmukh

0*0¶V'U*@ 0LUFR .RFKHU ³UniNE at CLEF 2015: Author 3URILOLQJ´1RWHERRNIRU3$1DW&/() [4] [4] Yasen Kiprov, Momchil Hardalov, Preslav Nakov, and ,YDQ .R\FKHY ³68#3$1  ([SHULPHQWV LQ $XWKRU 3URIOLQJ´ [5] [5] Jennifer Golbeck, Cristina Robles, Michon Edmondson, DQG .DUHQ 7XUQHU ´3UHGLFWLQJ 3HUVRQDOLW\ IURP 7ZLWWHU´ 2011 IEEE International Conference on Privacy, Security, Risk, and Trust, and IEEE International Conference on Social Computing [6] [6] $OH[DQGUD 5RVKFKLQD -RKQ &DUGLII 3DROR 5RVVR  ³ Comparative Evaluation of Personality Estimation AlgoritKP IRU 7:,1 5HFRPPHQGHU 6\VWHP´ https://en.wikipedia.org/wiki/Big_Five_personality_traits

315

Cognitive Knowledge Engineering

Chapter 65

Mining Movie Intention Using Bayes and Maximum Entropy Classifiers Varsha D. Jadhav

Sachin N. Deshmukh

P.E.S. College of Engineering Aurangabad, (MS) India

Department of CS/IT, Dr. B.A.M.U Aurangabad, (MS) India

[email protected]

[email protected]

ABSTRACT Sentiment analysis is becoming one of the most thoughtful research areas for prediction and classification. In this paper we try to analyze and predict the result for movie reviews. We use machine learning techniques Bayes and Maximum entropy for classifying text messages. We use movie comments from twitter. :HDQDO\]HWKHWZRFODVVLILHUVIRUKLQGLPRYLHVµ6XOWDQ¶ DQG µ0DGDDUL¶ :H UHWULHYHG WKH WZHHWV EHIRUH DQG DIWHU WKH release of the movie. We evaluate the accuracy and compare the Bayes and Maximum entropy method. We use R technology for the movie review analysis.

General Terms Intention Mining, machine learning, normalization, Twitter AP

Keywords Bayes, Maximum Entropy, polarity, emotions, mean absolute error.

1. INTRODUCTION Social networks today contains enormous amount of text data, which is growing everyday. Intention mining aims to cover the attitude of the author on a particular topic from text data. It is natural language processing and machine learning techniques reveal the attitude. In the recent years it has gained popularity due to its immediate application in business, customer feedback from product reviews, spots reviews and assisting in election campaigns. Movie reviews are a important way to analyze the performance of a movie. Text movie reviews tells us the strong and weak points of the movie which tells us whether the movie in general meets the expectations of the reviewer. Using intention mining, we can find the state of mind of the reviewer and understand the polarity and emotions. Polarity is positive, negative and neutral. Emotions are anger, disgust, fear, joy, sadness, surprise, and unknown. In this paper we use intention mining on a set of movie reviews extracted from twitter and try to understand the overall reaction about the movie. Whether people liked or they disliked it. We analyzed the movie reviews using machine learning methods Bayes and maximum entropy. We compare the two methods for accuracy.

presented the Bayes model to predict the intentions of the people who tweet on twitter about a specific topic. They predicted the cricket match result which serves as strategic guidance to the captains so as to improve the performance of the team. Kamal Nigam et.al [3] proposes the use of maximum entropy techniques for text classification. In their text classification scenario, maximum entropy estimates the conditional distribution of the class label given a document. A document is represented by a set of word count features. The labeled training data is used to estimate the expected value of these word counts on a class-by-class basis. Improved iterative scaling finds a text classifier of an exponential form that is consistent with the constraints from the labeled data. Kuat Yessenov, et.al [4], presented an empirical study of efficacy of machine learning techniques in classifying text messages by semantic meaning. They used movie review comments from popular social network Digg as the data set and classify text by subjectivity/objectivity and negative/positive attitude. Changlin Ma, et.al [5], proposed a novel topic and sentiment unification maximum entropy LDA model in this paper for fine-grained opinion mining of online reviews. Oaindrila Das, et.al [6] presented a novel approach for classification of online movie reviews using parts of speech and machine learning algorithms. Borislav Kapukaranov and Preslav Nakov [7] presented experiments in predicting fine- rained stars, including halves, for Bulgarian movie reviews. This is a challenging task, that can be seen as (a) multi-way classification, i.e., choosing one out of eleven classes, (b) regression, i.e., predicting a real number, or (c) something in between, namely ordinal regression, i.e., predicting eleven values, but taking ordering into account, e.g., predicting 4 when the actual value is 3.5 would be better than predicting 1.

3. METHODOLOGY Fig. 1 shows the framework for intention mining system.

2. RELATED WORK Minqing Hu and Bing Liu [1], studied the problem of generating feature-based summaries of customer reviews of products sold online. Here, features broadly mean product features (or attributes) and functions. Given a set of customer reviews of a particular product,the task involves three subtasks: (1) dentifying features of the product that customers have expressed their opinions on (called product features); (2) for each feature, identifying review sentences that give positive or negative opinions; and (3) producing a summary using the discovered information. Varsha D. Jadhav and S.N. Deshmukh [2]

Fig.1 Framework for Intention Mining

316

Natural Language Processing

The movie tweets are extracted using a keyword. The tweets are pre-processed which are classified for polarity and emotions using Bayes method and maximum entropy method. The intention is mined and decision is made.

4.1 Stemming is the process of reducing related word to the root word. E.g. generalization is generally represented as general.

Case Normalization

4. TWITTER API Twitter API is a twitter platform which connects the application with the world conversation happening on twitter. Once the twitter API is created we need to obtain credentials such as API keys, API secret, Access token and Access token secrete on the twitter developer site to access the twitter API. Twitter authentication is setup using these credentials. Using a keyword, tweets related to the specified keyword and language are searched. The language parameter that we have specified is English. The twitter library lets the use of Twitter API. The tweets retrieved are saved in .csv (comma separated file).

English text contains both higher and lower case characters. This process converts the complete sentences in lowercase or uppercase.

Usernames Removes @ symbol. Users often include twitter usernames by using @ symbol.

Usage links Retweeting is copying another users tweet and posting to DQRWKHUXVHUVDFFRXQW,WLVDEEUHYLDWHGZLWK³57´$OOWZHHWV with RT are removed.

5. DATA COLLECTION The tweets for the movie reviews were collected using the NH\ZRUG µ6XOWDQ PRYLH¶ XVLQJ 7ZLWWHU $3, IURP  th July 11, 2016 to 9th July 2016. Total 34,694 tweets were retrieved for polarity and emotion intention. The tweets collected each day for the movie Sultan are shown in table 1. The movie Sultan was released on 6th July 2016. Also the tweets for the movie rHYLHZV XVLQJ NH\ZRUG µ0DGDDUL PRYLH¶ ZDV FROOHFWHG IURP 20th July to 24th July 2016. Total 8033 tweets were retrieved. Table 2 shows the tweets collected on each day for the movie Madaari. The movie Madaari was released on 22th July 2016. Table 1. Daywise tweet coOOHFWLRQIRUWKHPRYLHµ6XOWDQ¶ Date 5th July 2016 6th July 2016 7th July 2016 8th July 2016 9th July 2016 Total

Number of tweets collected 6723 6997 6990 6993 6991 34,694

7DEOH'D\ZLVHWZHHWFROOHFWLRQIRUWKHPRYLHµ0DGDDUL¶ Date 20th July 2016 21th July 2016 22th July 2016 23th July 2016 24th July 2016 Total

Number of tweets collected 514 736 1518 2233 3032 8033

5.1 Data Preprocessing The raw data is pre-processed. The pre-processing includes

5.1.1 Tokenization Text data is collection of sentences which is split into terms or tokens by removing white spaces, commas and other symbols etc.

5.1.1 Stopword removal Removes articles such as (a, an, the).

Usage of N-gram and part of speech (POS) tags An n-gram is a N-character slice. N-gram of several different lengths are used simultaneously. POS feature is used because depending on the usage the same word may have different meaning.

5. BAYES METHOD We predict whether a review is negative, positive or neutral given only the text. In order to do this, we will train an algorithm using the reviews and classify to make predictions on the reviews. We will then be able to calculate our error and see how good our predictions were. For our classification algorithm, we used Bayes classifier. A Bayes classifier works by figuring out the probability of different attributes of the data being associated with a certain class. This is based on Bayes theorem. The theorem is

P( B | A) P( A) P( B) 6. MAXIMUM ENTROPY METHOD P( A | B)

The maximum entropy classifier is a discriminative classifier commonly used in natural processing and information retrieval problems. The Max Entropy classifier is a probabilistic classifier which belongs to the class of exponential models. Unlike the Naïve Bayes classifier, the Max entropy does not assume that the features are conditionally independent of each other. It is based on the principle of maximum entropy and from all the models that fit the training data, selects the one which has the largest entropy. The Entropy can be used to solve a large variety of text classification problems such as language detection, topic classification, sentiment analysis and more. Principle of MaxEnt states, take precisely stated prior data or testable information about probability function. Consider the set of all trial probability distribution that would encode the prior data. Of those the one with maximal information entropy is the proper distribution. The model is represented as

P (c | d , O )

1 exp[ ¦ i Oi f i (c, d )] Z (d )

P(c|d) is the probability that a class occurs for a given tweet, c is

O is a weight vector. The the class, d is the tweet, and weight vector decide the weight vector decide the significance of the feature in classification. Z(d) is a normalization function, fi is a feature class function. 317

Cognitive Knowledge Engineering

MAE tells us how big of an error we can expect from the forecast on average. Cort J. Willmott et.al [8] indicated that MAE is the most natural measure of average error magnitude than RMSE.

7. NORMALIZATION Data should be normalized to bring all the variables into proportion with one another when doing comparison analysis. The aim of normalization is to make variables comparable to each other. Normalization means to transform observations x into f(x) such that they look normally distributed. It simply means putting different variables on common scale. As seen from table 1. The number of tweets collected each day is different. The number of positive, negative and neutral polarities each day is different so we need to normalize it.

9. RESULTS AND DISCUSSION The results are obtained for Bayes and Maximum Entropy classifiers. Movie reviews of two movies, Sultan and Madaari were retrieved of which polarity and emotion intentions were obtained. The daywise polarity intention and the daywise emotion intention are obtained.

8. ACCURACY MEASURES

Table 3. Shows the daywise intention polarity using Bayes and MAXEN7 PHWKRGV IRU WKH PRYLH µ6XOWDn¶ The polarity is positive, negative and neutral.

The accuracy measures such as i. Mean error (ME) ii. Root mean square error (RMSE) iii. Mean absolute error (MAE) iv. Mean Percentage Error (MPE) v. Mean Absolute Percentage Error (MAPE) are used to measure the accuracy. MAE is simply the mean of the absolute errors. The absolute error is the absolute value of the difference between the forecasted value and the actual value.

The graph for daywise polarity intention using Bayes and MAXENT is shown in Fig.2. Using the graph the polarity intention can be easily seen and the polarity intention for positive, negative and neutral for both the classifiers, that is Bayes and MAXENT can be compared. The numbers of positive intentions are more than that of the negative and neutral intentions each day.

Table 3 .Daywise intention polarity using Bayes and MAXENT metKRGVIRUWKHPRYLHµ6XOWDQ¶ 5th July, 2016

6th July, 2016

7th July, 2016

8th July, 2016

9th July, 2016

Polarity Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Positive

4874

2699

3774

2664

4873

3898

4959

3797

4957

3728

Negative

1043

1017

2304

2233

1212

1093

1048

908

1087

973

Neutral

806

3007

919

2100

905

1999

986

2288

947

2290

Total

6723

6723

6997

6997

6990

6990

6993

6993

6991

6991

Fig. 2. Daywise Intention Polarity for the movie Sultan .

318

Natural Language Processing

The positive LQWHQWLRQSRODULW\ IRU µ6XOWDQ¶ IRU %D\HV method is more than the maximum entropy method each day.Table 4, Shows the day wise intention emotions using %D\HVDQG0$;(17PHWKRGVIRUWKHPRYLHµ6XOWDQ¶

Using Bayes method we can see that the number of joy emotions is more. Figure 3 shows the intention emotions for the movie Sultan. Unknown emotion is the limitation of our research. The number of joy emotion is the highest

Table 4. Daywise intention emotions using Bayes and MAXEN7PHWKRGVIRUWKHPRYLHµ6XOWDQ¶ 5th July, 2016

6th July, 2016

7th July, 2016

8th July, 2016

9th July, 2016

Emotions Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Anger

75

5347

139

5066

160

4601

60

4515

128

4629

Disgust

9

0

3

0

0

0

3

0

6

0

Fear

44

44

46

46

28

28

13

13

6

6

Joy

1081

1081

1462

1462

1625

1625

1980

1980

1618

1618

Sadness

160

160

114

114

52

52

137

137

251

251

Surprise

82

82

306

306

684

684

345

345

481

481

Unknown

5272

9

4927

3

4441

0

4455

3

4501

6

Total

6723

6723

6997

6997

6990

6990

6993

6993

6991

6991

Fig 3. Daywise Intention Emotion for the movie Sultan

319

Cognitive Knowledge Engineering

Table 5 shows the Mean Absolute Error and the accuracy using Bayes and MAXENT methods. Figure 4 shows the Accuracy for Bayes and MAXENT methods

10. NORMALIZATION OF DATA FOR THE MOVIE SULTAN 10.1 Bayes Method

Table 5.Mean absolute error and accuracy using Bayes and MAXENT for the movie Sultan

Let P1 be the sum of all positive intentions.

Date

Mean Absolute Error

% Accuracy

=

Bayes

MAXENT

Bayes

MAXENT

5 July 2016

2.6801

3.0515

97.3198

96.9484

6th July 2016

3.5256

4.0404

96.4743

95.9595

7th July 2016

3.2571

3.8182

96.7428

96.1817

8th July 2016

2.7058

3.0779

97.2941

96.9220

th

th

9 July 2016

2.9587

3.3732

P1 = 23437 Normalization factor for positive intention

97.0412

96.6267

23437 34697

= 0.6754

(1)

Let N1 be the sum of all Negative intentions. N1 = 6694 Normalization factor for Negative intention =

6694 34697

= 0.1929

(2)

Let NE1 be the sum of all Neutral intentions. NE1 = 4563 Normalization factor for Negative intention =

4563 34697

= 0.1315

(3)

10.2 Maxet Method Let P2 be the sum of all positive intentions. P2 = 16786 Normalization factor for positive intention =

16786 34697

= 0.4837

(4)

Let N2 be the sum of all Negative intentions. N2 = 6224 Normalization factor for Negative intention =

6224 34697

= 0.1793

(5)

Let NE2 be the sum of all Neutral intentions. NE2 = 11684 Normalization factor for Negative intention =

Fig. 4 Accuracy for Bayes and MAXENT for the movie Sultan The accuracy for Bayes method is more on each day as compared to maximum entropy. But the number of tweets collected each day is different. So the data needs to be normalized. The positive, negative and neutral polarities obtained needs to be normalized. Normalization is performed by adding all the positive polarities obtained on each day and divide it by the total number of tweets retrieved.

11684 34697

= 0.3367

(6)

The normalization factor for positive intention of Bayes method is more than the normalization factors of negative and neutral intention of the Bayes method. Also the normalization factor of positive intention of the MAXENT method is more than the normalization factor of negative and neutral intention of the MAXENT method. When comparing Bayes and MAXENT the normalization factor of positive intention of Bayes method is highest than that of the positive intention of MAXENT method. Table 6 shows the average of positive tweets for the movie Sultan

Normalization is performed for Bayes and MAXENT methods and then compared to get the best results.

320

Natural Language Processing

Table 6. Average for positive tweets for Bayes and MAXENT methods for Sultan Bayes

MAXENT

Date

No. of Positive tweets

Average

No. of Positive tweets

Average

5th July 2016

4874

72.4973

2699

40.1457

6th July 2016

3774

53.9374

2664

38.0734

7th July 2016

4873

69.7138

3898

55.7653

8th July 2016

4959

70.9137

3797

54.2971

9th July 2016

4957

70.9054

3728

53.3257

Total

23,437

337.9678

16,786

241.6074

more than that of negative and neutral intentions for Bayes approach. For the maximum entropy method the numbers of neutral intentions were more before the movie was released. The movie was released on 22nd July 2016. On 22nd,23rd and24th July 2016 the number of positive intentions are more. While comparing Bayes and maximum entropy methods the number of positive intentions obtained by Bayes method is more than that of the positive intentions obtained by maximum entropy method. Table 8 shows the intention emotions for the movie Madaari, and figure 6 shows the graph for the intention emotions. For Bayes method the numbers of joy emotions are highest. Unknown is the limitation of our research. For MAXENT method the numbers of anger intentions are more. But on the contrary the positive intention polarity if more, so we neglect it and consider other emotions. Making this assumption, the numbers of joy emotions are highest.

The Accuracy for all the 5 days is more for Bayes method. Considering the Bayes method alone, the numbers of positive tweets are more than negative and neutral tweets. Also, considering MAXENT method alone the numbers of positive tweets are more than that of negative and neutral tweets. Making comparison of Bayes and MAXENT the number of positive tweets for Bayes method are more than the number of positive tweets of MAXENT method every. Also the average of positive for Bayes and MAXENT methods for each day is calculated, as shown in table 6. Again calculating the average of total average for both Bayes and MAXENT it gives 67.5935% for Bayes method and 48.3214% for MAXENT method. Table 7 shows the daywise intention polarity for the movie µ0DGDDUL¶ )LJXUH  VKRZV WKH JUDSK IRU WKH GD\ZLVH SRODULW\ As seen from the graph the numbers of positive intentions are

Table 7. Daywise intention polarity using Bayes and MAXENT PHWKRGVIRUWKHPRYLHµ0DGDDUL¶ 20th July, 2016

21th July, 2016

22th July, 2016

23th July, 2016

24th July, 2016

Polarity Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Positive

265

149

394

232

929

601

1420

947

1956

1379

Negative

191

80

248

132

431

285

552

396

664

578

Neutral

58

285

94

372

158

632

261

890

412

1075

Total

514

514

736

736

1518

1518

2233

2233

3032

3032

321

Cognitive Knowledge Engineering

)L)LJ7ZLWWHULQWHQWLRQSRODULW\IRUWKHPRYLHµ0DGDDUL¶ Table. 8. Daywise Intention Emotions for the movie Madaari 20th July, 2016

21th July, 2016

22th July, 2016

23th July, 2016

24th July, 2016

Emotions Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Bayes

MAXENT

Anger

13

343

13

534

19

1140

32

1680

46

3217

Disgust

18

0

0

0

0

0

0

0

0

0

Fear

0

0

0

0

1

1

33

33

37

37

Joy

0

18

33

33

145

145

233

233

343

343

Sadness

152

152

162

162

178

178

197

197

187

187

Surprise

1

1

7

7

54

54

90

90

148

148

Unknown

330

0

521

0

1121

0

16487

0

2271

0

Total

514

514

736

736

1518

1518

2233

2233

3032

3032

322

Natural Language Processing

Fig.6. Daywise Intention Emotion for the movie Madaari Table 9 shows the Mean Absolute Error and the accuracy using Bayes and MAXENT methods IRUWKHPRYLHµ0DGDDUL¶ Figure 7 shows the graph of Accuracy for Bayes and MAXENT methods.

The accuracy for Bayes method is more each day as compared to that of the Maximum Entropy method.

Table 9. Mean absolute error and accuracy using Bayes and MAXENT for the movie Madaari

Date

Mean Absolute Error

% Accuracy

Bayes

MAXENT

Bayes

MAXENT

20th July 2016

2.0381

2.4320

97.961

97.5679

21th July 2016

1.9463

2.3186

98.0536

97.6813

22th July 2016

1.9486

2.2427

98.0513

97.5727

23th July 2016

1.9440

2.2383 98.0559

98.0559

97.7616

24th July 2016

1.9496

2.2469

98.0503

97.7530

Fig. 7 Accuracy for Bayes and MAXENT for the movie Madaari The number of tweets collected each day is different. The data needs to be normalized for each day as well as for the positive, negative and neutral polarities. The total numbers of tweets collected for the movie Madaari are 8033.

323

Cognitive Knowledge Engineering

Normalization factor for positive intention

11. Normalization of data for the movie Madaari

=

11.1 For Bayes Method

1379 = 0.1716 8033

(10)

Let P3 be the sum of all positive intentions. Let N2 be the sum of all Negative intentions. N4 = 578 Normalization factor for Negative intention

P3 = 4955 Normalization factor for positive intention =

4955 8033

= 0.6168

(7)

=

Let N3 be the sum of all Negative intentions.

2068 8033

= 0.2574

(8)

=

Let NE1 be the sum of all Neutral intentions.

683 = 0.0850 8033

1075 = 0.1338 8033

(12)

Considering Bayes method, the normalization factor for positive intention is more than the negative and neutral intentions. Also, for MAXENT method, the normalization factor for positive intention is more than that of the negative and neutral intentions. Comparing Bayes and MAXENT the normalization factor of positive intention of bayes method is highest than the normalization factor of positive intention of MAXENT method. Table 10 shows the average of positive tweets for the movie Sultan

NE3 = 683 Normalization factor for Negative intention =

(11)

Let NE2 be the sum of all Neutral intentions. NE4 = 1075 Normalization factor for Negative intention

N3 = 2068 Normalization factor for Negative intention =

578 = 0.0719 8033

(9)

11.2 For MAXENT Method Let P2 be the sum of all positive intentions. P4 = 1379

Table 10. Average for positive tweets for Bayes and MAXENT methods for Madaari Bayes

MAXENT

Date

No. of Positive tweets

Average

No. of Positive tweets

Average

20th July 2016

256

49.8054

149

28.9883

21th July 2016

394

53.5326

232

31.5217

22th July 2016

929

61.1989

601

39.5915

23th July 2016

1420

63.5915

947

42.4093

24th July 2016

1956

64.5118

1379

45.4815

Total

4955

292.6402

3308

187.9923

The accuracy for Bayes method is more than the accuracy of MAXENT method. Again calculating average of the total average of both Bayes and MAXENT methods, we get 67.5534% for Bayes method and 48.3830% for MAXENT method.

12. CONCLUSIONS In comparison of machine learning classifiers, Bayes and Maximum Entropy, mining intention for movie review prediction, the Bayes classifier is more efficient than the MAXENT. We have obtained 67.5935% accuracy for movie µ6XOWDQ¶ XVLQJ %D\HV PHWKRG DQG  XVLQJ 0$;(17 method. The movie Sultan was released on 6th July 2016. According to the Hindi movie reviews of Times of India dated 6th July 2016 the review rating is 3.5 out of 5, which is 70%.



Our prediction is close to the Hindi movie reviews of Times of India. For the movie Madaari Bayes classifier is giving 67.5935% and MAXENT classifier is giving 48.3830% accuracy. The movie Madaari was released on 22nd July 2016. According to the Hindi movie reviews of Times of India dated 22nd July 2016 the review rating is 3.5 out of 5, which is 70%. Our prediction using Bayes classifier for the movie Madaari is close to the Hindi movie reviews of Times of India. Also the normalization factor for the positive intention for Bayes method is highest than that of MAXENT method for both Sultan and Madaari movies. It is 0.6754 for Sultan movie and 0.6168 for Madaari movie. Hence, we can conclude that Bayes method is efficient than Maximum entropy method for text classification.

324

Natural Language Processing

13. FUTURE WORK More machine learning classifiers can be used to analyze text data. Also anger and unknown emotions can be further analyzed to get more accurate results.

14. ACKNOWLEDGMENTS The authors are deeply thankful to the Department of Computer Science and Information Technology, Dr. Babasaheb Ambedkar Marathwada University Aurangabad. The authors are also thankful to Prof. R.R. Deshmukh, Head of department of Computer Science and Information Technology, Dr. Babasaheb Ambedkar Marathwada University Aurangabad. We express our Gratitude to Dr. Abhijeet P. Wadekar, Principal, P.E.S. College of Engineering, Aurangabad, for the support during the research work.

15. REFERENCES [1] MinqinJ +X DQG %LQJ /LX ³Mining and Summarizing &XVWRPHU5HYLHZV´''¶, August 22±25, 2004, Seattle, Washington, USA. [2] Varsha D Jadhav, S.N. Deshmukh, ³7ZLWWHU ,QWHQWLRQ Classification Using Bayes Approach for Cricket Test 0DWFK 3OD\HG %HWZHHQ ,QGLD DQG 6RXWK $IULFD ´ Proceeding of International Conference on Internet of Things, Next Generation Network and Cloud Computing 2016. ISSN: 0975 ± 8887,

[3] Kamal Nigam,John Lafferty, Andrew McCallum, ³Using maximum entropy IRU WH[W FODVVLILFDWLRQ´ ,Q ,-&$,-99 Workshop on Machine Learning for Information Filtering. [4] Kuat Yessenov, Sasa Misailovic, Sentiment Analysis of 0RYLH 5HYLHZ &RPPHQWV´ 6.863 Spring 2009 final project, May 17, 2009. [5] Changlin Ma, Meng Wang, Xuewen &KHQ ³Topic and Sentiment Unification Maximum Entropy Model for Online 5HYLHZ $QDO\VLV´ WWW 2015 Companion, May 18±22, 2015, Florence, Italy.ACM 978-1-4503-34730/15/05. [6] Oaindrila Das, Rakesh Chandra Balabantaray ³Sentiment Analysis of Movie Reviews using POS tags and Term )UHTXHQFLHV´ International Journal of Computer Applications (0975 ± 8887) Volume 96± No.25, June 2014. [7] Borislav Kapukaranov, Preslav Nakov, ³)LQH-Grained Sentiment Analysis for 0RYLH5HYLHZVLQ%XOJDULDQ´ Proceedings of Recent Advances in Natural Language Processing, pages 266±274,Hissar, Bulgaria, Sep 7±9 2015. [8] Cort J. Willmott, Kenji Matsuura, ³Advantages of the mean absolute error (MAE) over the root mean square error RMSE) in assessing average model Performance, Climate Research, Vol. 30: 79±82. [9] http://timesofindia.indiatimes.com, Hindi movie reviews.

325

Cognitive Knowledge Engineering

Chapter 66

Conceptual Dependency for understanding question: QA system Kalpana Khandale

Hanumant R. Gite

C. Namrata Mahender

Dept of Comp Sci & IT Dr. B .A .M. University Aurangabad, MH (India)

Dept of Comp Sci & IT Dr. B .A .M. University Aurangabad, MH (India)

Dept of Comp Sci & IT Dr. B .A .M. University Aurangabad, MH (India)

[email protected]

[email protected]

[email protected]

ABSTRACT Question answering is a blooming field with all its fantasies the question answering system widely classified based on the way they access the knowledge i.e. web based or open domain and restricted or domain specific as closed domain. Question answering if deployed at its full edge demonstrate a lot application ranging from HCI to search engine, medical aspect to security issues. The major hurdle that is discussed in this paper is the method through which the question can be made understood that too with few rules. Our work focuses on factoid base questions which are framed based on Blooms and Lenhert taxonomy. This paper presents on idea of how CD has helped in some sense to extract the conceptual need of a given question and extract the level of its understanding or response requirement

Keywords Natural Language Processing, Question, Answer, QA System, Conceptual Dependency

1. INTRODUCTION 1.1 Question Answering System: Research in the area of Question Answering generates a lot of interest both from the NLP community and from the end-users of this technology, either lay users or professional information analysts. Research needs to be done to expand and consolidate these categories for the larger scope of question answering. The development of QA from early procedures that mapped natural language questions into queries in a standard database language for a closed data set, to contemporary open systems that seek answers to questions across a large set of documents, often the entire web.QA system can consider by adding a set of semantic implications to a question and its range of possible answers. [1] As the ultimate goal of the Q&A paradigm is the creation of an integrated suite of analytical tools, the Q&A systems should be extended to dialogue processing systems that are queried and respond in real time to problems. There are two types of Question Answering System a) Open-domain QA System [2] and b) Closed-domain QA System. [3] ,QRUGHUWRILQGDFRUUHFWDQVZHUWRDXVHU¶VTXHVWLRQZHQHHGWR first know what to look for in our large collection of documents. The type of answer required is related to the form of the question, so knowing the type of a question can provide constraints on what constitutes relevant data, which helps other modules to correctly locate and verify an answer. The question type classification component is therefore a useful, if not essential component in a QA system, as it provides significant guidance about the nature of the required answer.

Now-a-days the QA system is most important for the vast variety of the text document. Many researchers search the many techniques for the development of the text analysis, question understanding, finding exact answer and many more.

1.2

Major component of QA:

1.2.1 Question Analysis In the question analysis module, questions posed to the system are processed to detect and extract information that might be useful to the other modules. This is carried out by two main tasks: (1) a classification of the question to determine the type of information that the question expects as answer (a date, a quantity, etc.), and (2) the selection of those elements that will allow the system to locate the documents that are likely to contain the answer.

1.2.2 Question Classification An important part of question analysis is the classification of the question into categories that represent the type of answer expected. The question in what year did India elect its first female prime minister? Seeks a date as an answer, for example, whereas the question How far is Delhi from Mumbai? Seeks a distance as an answer, the classification of the answer type in some questions is complicated by the fact that there might be ambiguity in the type of expected answer. Thus, the question how long does it take to travel from India to London?, Can accept either a time or a distance as an answer [4]. The identification of the answer type is usually performed by analyzing the interrogative terms of the question (wh-terms). For example, given the question who is she? The term where indicates that the question is looking for a person [4].

1.2.3 Query Generation Question-analysis also extracts the information in the question that allows the generation of queries that can be processed by an IR system, which will facilitate the selection of the answerbearing text extracts from the entire document collection. These queries are commonly obtained using one of two different processes: keyword selection and answer-pattern generation [4].

1.3 Document passage selection The document or passage selection module uses part of the information extracted during the question analysis module to perform an initial selection of answer-bearing candidate texts. Given the great volume of documents these systems are expected to manage, and the limitations imposed by a

326

Natural Language Processing

reasonable response time, this task is carried out using IR or paragraph retrieval (PR) systems.

1.4 Answer Extraction The answer extraction module carries out a detailed analysis of the relevant texts selected by the document selection module in order to locate and extract the answer. To achieve this, the representation of the question and the representation of relevant texts are matched against each other in order to obtain a set of candidate answers.

1.5 Types of QA: 1.5.1 Open Domain Question Answering:Open domain question answering deals with questions about nearly everything and can only rely on general ontology and world knowledge. [5]

1.5.2 Closed Domain Question Answering System:Closed-domain question answering deals with questions under a specific domain (for example medicine or weather forecasting, etc) [6]

2. PROPOSED QUESTION ANSWERING MODEL For evaluating or accessing candidate answer there is a need of a query processor. An Inference Engine is developed to address the issues in our question answering model is show in Fig 2. The inference engine has majorly divided into three components as generic inference engine 1) Repository & knowledge base 2) Query processor 3) user interface.

2.1 Repository & Knowledge Base The repository is the collection of all answers collected from each user. While knowledge base unit contains separately the model questions and model answers.

3. COMPONENTS OF OUR SYSTEM 3.1 Preparation of Tutorial Before preparing tutorial there is need to understand what is tutorial? A tutorial is a method of transferring knowledge and may be used as a part of a learning process. More interactive and specific than a book or a lecture; a tutorial seeks to teach by example and supply the information to complete a certain task [8]. By keeping this point in the picture, we tried to prepare LQIRUPDWLYHEXWVPDOOWXWRULDOVSOXVDQDUWLFOHIURPWKHUHDGHU¶V digest September 2009 was considered as tutorial, to check whether it is possible to answer both tutorials in the same manner. In these work we consider one tutorial on article on Stress & Health from reader digest September 2009 was prepared.

3.2 Preparation of Question Paper Question designing also requires a lot of skill, as this helps us to find out how much the subject or concept are understood by the candidate, are they able to apply them in real time situation etc. On the basis of tutorials we prepared question paper that includes objective and subjective question.

3.3 Preparation of Model Answer The answers were prepared by experts, taking consideration of the tutorial it was observed if more than one way of answering the same question was possible then all those were considered as part of answer repository. Main aspect of whole system relies on question understanding. The section below discusses about question conceptual understanding of question.

4. QUESTION Question, the word has never needed a dictionary to know, but the response needs not only a dictionary, but the whole universe sometimes to be searched for just a response. The question is a QRXQ ³$ VHQWHQFH UHFRUGHG RU H[SUHVVHG VR DV WKH HOLFLW LQIRUPDWLRQ´ In general every question has a response but every question response may be dealt in different ways, like few questions have answers which are of fix type. For example: What is your name? Answers:

a. my name is Ram.

But such responses can be paraphrased as following. b. Ram. c. Ram is my name.

Fig. 1. Architecture of Inference Engine

2.2 Query Processor

d. All call me as Ram.

It takes the answer from the user interface, frames the required query for processing.

e. I am called as Ram.

2.3 User Interface It provides an interface for interacting with the inference engine. The inference model fetches the candidate answer in the processing unit & extracts its relative answers from the model answer & by matching process the production unit decides about the correctness of the answer. The design of the production system is rule based. The questions and their model answer are stored in the plain text format in the respective repository [7].

There are certain questions which may have varying responses i.e. is meaning of responses may also we varying. E.g.:- Do you have watch? Answer:

a. Yes, I have. E1R,GRQ¶WKDYH c. The time is not good for me. d. Now the time is 6:00 clock.

E.g.:- What time it is? 327

Cognitive Knowledge Engineering

Answer:

a. 10 am. b. Good time c. Bad time

It may be context with asking about or referring a new watch. That states that many times our responses are context based. Answers based con-text requires the knowledge and understanding of the given content while paraphrased answers commonly should tend to one concept.

4.1 Need of question Taxonomy Question answering techniques are developed considering the possibilities offered by the different types of questions that can be asked to the system, the targeted source for answers, and how answers are presented to the user.

5.

Synthesis

6.

Evaluation

4.1.1.2 W. Lenhert conceptual categories /HKQHUW¶V approach provides a means for categorizing questions on the basis of meaning and the type of response that is appropriate. [11] Table 2. Question Categories of W. Lenhert taxonomy Sr. No. 1. 2.

Question Categories Causal Antecedent Goal Orientation Enablement Casual Consequent Verification

There is need for developing Q&A systems that are capable of extracting answers from large data sets in several seconds, regardless of the complexity of the question, the size and multitude of the data sources or the ambiguity of the question.

3. 4.

It is often the case that the information need is not well captured by a QA system, as the question processing part may fail to classify properly the question or the information needed for extracting and generating the answer is not easily retrieved. In such cases, the questioner might want not only to reformulate the question, but to have a dialogue with the system.

6. 7.

9.

Disjunctive Instrumental/ Procedural Concept Completion Expectation

The taxonomy was developed to handle enquiries. In mechanism that generates genuine information seeking questions, so we need a taxonomy that emphasizes inquiries rather than interrogative expression per se. Moreover the taxonomy was not developed on the basis of a fortuitous sample of observations. The question categories on the basis of meaning rather that form are consistent with some other taxonomy of questions in the cognitive science.

10.

Judgmental

11.

Quantification

More sophisticated questioners expect answers that are outside the scope of written texts or structured databases. To upgrade a QA system with such capabilities, it would be necessary to integrate reasoning components operating on a variety of knowledge bases, encoding world knowledge and commonsense reasoning mechanisms, as well as knowledge specific to a variety of domains. [9]

4.1.1 Question Type 4.1.1.1 Blooms Taxonomy: Cognitive is the most-used of the domains by blooms taxonomy of types of Question in education. It can be viewed as a sequence of progressive contextualisation of the material. [10]

Table 1. Question Categories of Blooms Taxonomy

1. 2.

Question Categories Knowledge Comprehension

3.

Application

4.

Analysis

Sr. No.

Question Examples Who wrote Wings of Fire? What is the main idea of this story? What happens when you multiply each of these numbers by nine? Why did the United States go to

war with England? How would your life be different if you could breathe under water? What story did you like the best?

5.

8.

Examples Why did Magy go to London? How did the glass break? For what purpose did Magy take the book? How was Magy able to eat? What happened when Magy left? Did Magy leave? Does Magy think that John left? Was Magy or John here? How did Magy go to London? How do I get to your house? What did Magy eat? Who gave Magy the book? :K\GLGQ¶W0DJ\JRWR/RQGRQ" :K\LVQ¶W0DJ\HDWLQJ" What should Magy do to keep Magy from leaving? How many people are there? How many brothers does Magy have?

Taxonomy is also impacted by the questioner. According to John Burger following levels of questioner are:

4.1.2 Level of questioner [12] 4.1.2.1 Casual questioner: In this type of questioners normal questions are posing to the V\VWHP 0DMRUO\ LW IRFXV LQQRUPDO ³SHUVSHFWLYH´ WR KDQGOH WKH TXHVWLRQV OLNH (J³:KHQ KH ZDV FRPH"´ DQG ³ZKR LQYHQWHG WHOHVFRSH"´  $OO WKHVH W\SH RI questions are having normal context.

4.1.2.2 Template questioner: In this type of questioners, templates are generated for the given TXHVWLRQ ZKLFK IRFXVHV RQ WKH ³OLQJXLVWLF´ NQRZOHGJH RI WKH TXHVWLRQ)RU(J³+RZ5DPPDQDJHWRFRPSOHWHDZRUN"´DQG ³'RHVDQ\VSHFLILFUHDVRQWRLQYHQW4$V\VWHP"´

4.1.2.3 Cub Reporter: In this type of questioners the complex questions are broken down into small set of questions. It majorly consists of context and specific relations to answer the questions of this type. It can answer thH TXHVWLRQV OLNH (J ³'RHV DQ\ VSHFLILF DFWLRQV SHUIRUPHG E\ 86 JRYHUQPHQW DIWHU /LQFROQ¶V GHDWK"´ &XEH reporter generates small set of questions which are associated to WKH FKLHI TXHVWLRQV WKDW DUH (J   ³:KHQ GLG 6K\DP GLHG"´ ³:KDW ZDV WKH UHDVRQ EHKLQG KLV GHDWK"´ DQG ³:KDW ZDV UHOHDVHGE\,QGLDQJRYHUQPHQWDIWHU,QGLUD*DQGKL¶VGHDWK"´

328

Natural Language Processing

4.1.2.4 Professional Information Analyst: These questions are having future perspectives. It is used to identify different taxonomies and multiple facts which are involved in the questions, but it requires much reasoning WHFKQLTXHVIRUDQVZHULQJWKHTXHVWLRQVOLNH(J³:KDWDUHWKH actions done by Indian government to honor Mahatma *DQGKL"´-panel menu-selection schemes or To understand the concept or meaning or the requirement posed in a question, the first thing is to analyze them, extract the said information so that the appreciate response can be dealed. The process in it as lot of issues uses following are the few important ones, i. ii. iii. iv.

Proper representation To identify what to be extracted Extraction and analysis of the identified information Word meaning, sentence level meaning, pragmatic meaning to be deployed for better & quick understanding The databases of factoid based question were framed on a brief introduction on Wikipedia. Total 60 factoid based on wh-type as shown in the table 4 Table 3. Question Dataset of the wh-type question Sr. No. 1. 2. 3.

Question Type What What What

4.

What

5. 6. 7. 8. 9.

What What What What What

10 11 12

What What What

13 14 15 16 17 18 19 20 21

What What What What What What What What What

22

What

23 24 25

What What What

26 27 28 29

What Where Where How

30

How

31

How

Example What is Wikipedia? What is mean by Wikipedia? What type of information does contain Wikipedia? What is thinking about the distribution of free encyclopedia of a creator? What is encyclopedia? What is the problem of Wikipedia? What is operated by Wikipedia now? What is Wiki stands for? Give too what Wikipedia will always get bigger and better? What type of website is Wikipedia? What type of book is Wikipedia? What are the most interesting features of Wikipedia? What does Wikipedia allow to do? What do you mean by vandalism? What does the articles holds about itself? What it does Wikipedia foundation has? What is Wikipedia exist? What type of content Wikipedia is? What is meant by open content? What is meaning of Wiki? What is fundamental principle of Wikipedia? What is the continues updating of Wikipedia? What is dictionary of encyclopedia? What is the focus of it? What modern dictionary evolved of encyclopedia? What is the example of free encyclopedia? Where we can find/get Wikipedia? Where is the Wikipedia available? How many editions in language on Wikipedia? How many entries of languages one there on Wikipedia? How many articles are there on English

32

How

33 34 35 36

How How How How

37

How

38

How

39

How

40 41 42 43

How How How How

44 45 46 47 48

Who Who Who Who Who

49 50 51

Who Who Why

52 53 54 55 56

Why Why Which Which Which

57

Which

58

Which

59 60

When Do

language? How much cost to be invested for Wikipedia? How the Wikipedia is written? How many people contribute for it? How many articles are added everyday? How many words combination does Wikipedia stands? How do the writers contribute to Wikipedia? How do we get information through Wikipedia? How is Wikipedia different than an encyclopedia? How the encyclopedias work as book? How encyclopedias divide? How many years existed encyclopedias? How we grow the knowledge with encyclopedia? Who writes articles? Who can make changes in the article? Who is the creator of Wikipedia? Who are the writers? Who is responsible for the articles on Wikipedia? Who is founded as an offshoot? Who are the readers of the Wikipedia? :K\WKH:LNLSHGLDFDOOHGµJRRG-natured ZHEVLWH¶" Why is Wikipedia special website? Why it is called as book of knowledge? Which foundation operates Wikipedia? Which aged people edit the Wikipedia? Which company support to free encyclopedia? Which century encyclopedia creates a dictionary? In which form does Wikipedia hold or contain the article? When Wikipedia launched? Do the have to pay the Wikipedia?

These questions were represented using conceptual dependency. The following table 4 shows CD representation. During the strategies on question taxonomy we found that the Blooms taxonomy categories the question on the basis of what type of response is need while the Lenhert focuses on the context association of the topic or relevance with prior information. To represent this information for acknowledging what should be extracted based on information gained from the question; the questions are represented in graphical form using conceptual dependency and the extracted information is accessed for question classification in our work. The next section discuss in detail the exploration and analysis on it.

5. CONCEPTUAL DEPENDENCY REPRESENT QUESTION CONCEPT

TO

The linguistic process can be thought of, in Conceptual Dependency terms, as a mapping into and out of some mental representation. This mental representation consists of concepts related to each other by various meaning-contingent dependency links. Each concept in the inter-lingual network may be associated with some word that is its realization on a sentential level.

329

Cognitive Knowledge Engineering

Now-a-days the lack of satisfactory systems that provide questions in domain which are logically complex and poorly structured. A major goal of this work is that is attempt to analyze natural language into meaning (conceptual) structures that are unambiguous representations of the meaning of an input

utterance. With the conceptual dependency representation we represent a question for understanding. For representing in CD 15 general syntactic rules are provided those are categorized based lexical level and conceptual level. Table 2 shows the CD representation of few sample questions from the database. [13]

Table 4. The syntax rule of the conceptual dependency for sample Questions. Question Class

Example of Question

What asked / Answer

Difficulty Level

Graphical Representation in CD ?

What

What

What is Wikipedia?

What is encyclopedia?

Meaning or definition / (Medium) The free encyclopedia.

Wikipedia

MBUILD

Meaning / Set of book. (Medium)

Encyclopedia

o

information

D Wikipedia ?

MBUILD

o information D Encyclopedia Obj2

Obj1

ATRANS

o

Language

D Obj1

How

How many editions in Quantity / more than languages on Wikipedia? 285 languages.

Difficult (Simple)

? Obj1 Obj2

ATRANS

o

editions R Obj2

How Who Who

Which

Which

How many people contribute for the Wikipedia?

Money / More than 77,000 people contribute to it.

Medium

Who is the creator of the Name / Jimmy Wales Simple Wikipedia? Who are the writers of Writer / It is written by Wikipedia? all the people who want simple to contribute to it. Which type of articles Wikipedia contain?

Which foundation operates Wikipedia?

Type or format / tables, diagrams, pictures, Simple photographs and beautiful illustrations.

Wikipedia Obj1/?

o

Contributed D People

?

ATR ANS

o OWNERSHIP : Wikipedia

?

ATRANS

o articles

Wikipedia D Writers ? ?

ATRANS

o

articles

D Format

Name / the non-profit Simple Wikipedia-foundation

? ?

Where

ATRANS

Where can Location / world wide Simple we find/get Wikipedia? web (WWW).

ATRANS

o

Foundation

D Wikipedia ?

?

ATRANS

o

Wikipedia

D X(LOC)

Why

Why the Wikipedia FDOOHGDVµJRRGQDWXUHGZHEVLWH¶"

Features / Wikipedia is Complex a book of knowledge that will always get bigger and better

The representation helped us to investigate following concepts. i.

Question have direct association on wh-type

E.g. What is your name: direct association. ii.

? ?

MBUILD

First contribution aspect then total no. Thus simple and not direct. Some questions require reasoning, prior information

Website

D Wikipedia

(J:K\WKH:LNLSHGLDFDOOHGDVµJRRG-QDWXUHGZHEVLWH"¶ i.

List out what is the meaning of good-natured

ii.

Then implies information from the paragraph and associate it with good-natured.

iii.

Then entitled the association is valid.

Few questions has two parts

E.g. How many people contributed for the Wikipedia?

o

Thus such questions are complex in nature. The proposed method has provided a way to identify the extraction of concept by its representative Method [CD] .the experiment carried demonstration as explained in the table 3 330

Natural Language Processing

that depending on the topic cogent from we are able to clarify the efforts required for extracting the response for a given question. The analysis is briefly depicted in table no. 5. In above experiments the tenses expressed in the questions have not been considered , special attention is not been given to the pragmatic aspect type of question which is one of the limitation of the work.in future process the meaning ,and a position value will be considered along with the part of speech level for conceptualization of the question. Table 5 The Overall results based on CD representation No. of Questions

CD Type

Understand ing

25

ATRANS

Straight

35

MBUILD

20

Mix [ATRANS,PRO DEL,INGEST]

Indirect of relation based associated Straight

Level Type of question Simple

8. REFERENCES [1] Alexander Clark, Chris Fox, and Shalom Lappin, The Handbook of Computational Linguistics and Natural Language Processing a John Wiley & Sons, Ltd., Publication. [2] Mark Andrew Greenwood, Department of Computer Science University of Sheffield, UK September 2005, Open-Domain Question Answering. [3] Poonam Gupta ME Research Scholar, Vishal Gupta Assistant Professor, A Survey of Text Question Answering Techniques, Computer Science & Engg. University Institute of Engg and Tech. Panjab University, Chandigarh, India,

Complex [4] Simple to Medium

6. CONCLUSION Now-a-days the Question Answering System is a buzz word in natural language processing. Under it the open-domain question answering system is most popular and suitable for the open sources. With the help of open-domain question answering systems we can deals with the questions about everything and only rely on the general ontology and world knowledge. In the closed domain system it deals only with question under the specific purpose like examination, music, energy. In the research area of computer science there is lot of problem with the question understanding. There are many common answers for a question. The extraction of answer is depending on the complexity of the question. To understand question Blooms and Lenhert taxonomy has been referred in our work. Factoid based 60 questions are used for this experiments. Here, we have used the conceptual dependency representation for the question understanding and proper framing the concepts. The CD have helped to reduce 70% of efforts or in another way provided with unique way for representing the types of question associations. Still few limitations in CD avoids the question framing in accordance to tenses, reasons due to content knowledge and many more. In our further work we would like to exploit these limitations with a hybrid approach.

7. ACKNOWLEDGEMENT Authors would like to acknowledge and thanks to University Grants Commission (UGC), India for granting Rajiv Gandhi National Fellowship (RGNF) for supporting this work and Department of Computer Science and Information Technology, Dr. Babasaheb Ambedkar Marathwada University, Aurangabad, Maharashtra, India.

Boris Katz, Gary Borchardt and Sue Felshin Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory32 Vassar Street, Cambridge, MA 02139 START Natural Language Annotations for Question Answering

[5] Alexander Clark, Chris Fox, Shalom Lappin "The Handbook of Computational Linguistics and Natural Language Processing", Wiley Blackwell ISBN 978-14051-5581-6, 2010. [6] L. Hirschman, R. Gaizauskas "Natural Language question DQVZHULQJWKHYLHZIURPKHUH´1DWXUDO/DQJXDJH Engineering 7 (4): 275-300. 2001 Cambridge University Press, DOI: 10.1017/S1351324901002807 Printed in the United Kingdom. [7] Gite H.R., Dhokrat Asmita and Namrata C Mahender ³'HYHORSPHQW RI ,QIHUHQFH (QJLQH WR $XWRPDWH WKH 'HVFULSWLYH([DPLQDWLRQ6\VWHP´,QWHUQDWLRQDO-RXUQDORI Computer Applications 65(22):40-43, March 2013. Published by Foundation of Computer Science, New York, USA. [8] http://english.stackexchange.com/questions/48388/differen ces-between-tutorial-guide-and-how-to [9] %XUJHU -RKQ HW DO  ³,VVXHV 7DVNV DQG 3URJUDP Structures to Roadmap Research in Question & Answering 4 $ ´ XQSXEOLVKHG PDQXVFULSW http://wwwpir.nist.gov/projects/duc/papers/qa.Roadmap paper_v2.doc. [10] David R. Krathwohl,A, A Revision of Bloom's Taxonomy: An Overview, THEORY INTO PRACTICE,Volume 41,Number 4, Autumn 2002. [11] Questions and Information Systems By Thomas W. Lauer, Eileen Peacock, Arthur C. Graesser,Psychology Press, 15Apr-2013 [12] -RKQ%XUJHUHWDO,QIHGDFXN-DQ³,VVXHV7DVNV and Program Structures to Roadmap Research in Question $QVZHULQJ4 $ ´ [13] R. Schank, Conceptual Dependency: A theory of natural language understanding, Cognitive Psychology 3, 552-631 (1972)

331

Cognitive Knowledge Engineering

Chapter 67

Question Based Text Summarization under Rule Based Framework Deepali K. Gaikwad

Vaishali Kadam

C. Namrata Mahender

Department Of CS and IT, Dr. BAMU, Aurangabad Maharashtra, India

Department Of CS and IT, Dr. BAMU, Aurangabad Maharashtra, India

Department Of CS and IT, Dr. BAMU, Aurangabad Maharashtra, India

[email protected] om

vaishu7817kadam@yahoo .com

[email protected]

ABSTRACT Text summarization is important application of natural language processing. But not much work has been done in Marathi language. The current paper, represent the question based text summarization system for Marathi language with rule based framework. The rule based approach is used to generate question on Marathi text for this POS Tagger and NER (Named Entity Recognition) techniques are utilized for recognition of noun as the name or location from the given input text and generate the question on it.

Keywords Text summarization, Natural Language Processing (NLP), Question Based Text Summarization (QTS), Question Generation Process, Rule Based Approach

1. INTRODUCTION Text summarization is a process of collecting important information from original text and present in the form of summary. Various application of text summarization like HGXFDWLRQ ILHOG VRFLDO PHGLD QHZV DUWLFOH¶V WZLWWHU PDVVDJH facebook massages), biomedical field, government offices, researcher, etc. [1], Text summarization work has been done many languages namely foreign languages like English, French, Italian, Arabic, Spanish, Japanese, china etc., as well as Indian Languages like Hindi, Punjabi, Tamil, Telgu, Kannada, Bangali are available[2, 3, 4]. Automatic Text Summarization categories into two types: Single document summarization and multi- document summarization. Text summarization divided into two categories: Extractive and Abstractive text summarization. In extractive text summarization collect important text from original data and group them together without changing its meaning. It is proposed based POS tagging. The important sentence is based on statistical and linguistic feature of sentence. Abstractive summarization consists of understanding the source text by using linguistic method to interpret the text and expressing it in own language [2, 6, 7]. Abstractive text summarization classified into structured approach and semantic approach. Structured based approach extract important information from the document through such schemes as templates, extraction rules and other structures such as tree, ontology, lead and body phrase structure. And Semantic based approach, focused on semantic representation of document as well as identify noun phrase and verb phrase by linguistic data processing. The multimodal semantic model,

information item based method and semantic based methods are semantic based approaches. TF-IDF, cluster based, graph theoretic, machine learning, text summarization using fuzzy logic, neural network and LSA method these are the techniques of extractive text summarization [2, 7]. The paper presents a very new approach of summarization using Questions as guideline to pick the important aspect of the given text. The proposed method is to generate Question that accepts Marathi text as input and processes the input by applying POS tagging and NER then generate the question as per rules. The answer of the generated question is the summary of the given input but this paper limits its discussion only up to Question generation. Question based text summarization is discussed in detail next sections.

2. QUESTION BASED TEXT SUMMARIZATION For summarization the one thing which acts as a reminder is to get the main theme from the text. For performing these we took certain text and repetitively ask questions while doing these. We found that the answer can be clubbed together to form a brief answer on given text i.e., nothing but we can say summarization. So, we thought of auto generation of question of the text. The proposed work follows the following stages: 1. Acquiring the given text and splitting in into sentences. 2. Splitted sentences are passed to a POS tagger for word tagged information. 3. The nouns from each tagged sentences are further classified with noun referring to person and noun referring to location, other noun formation are not consider for this work. 4. The name entity especially person noun is replace with who and location is replace with where. 5. The replace questions are validated manually. For Example 1. ᮰֠ ᮧօ֎ ֐֡ովᱮ ֛֧ ֏֞֒ֆ֞ռ֧ ᳦֒֞֌֟ֆ ը֛֧ shri. Pranab Mukharji is President of India. Output: x ֏֞֒ֆ֞ռ֧ ᳦֒֞֌֟ֆշ֫օը֛֧"(Who is President of India?) x ᮧօ֎֐֡ովᱮշ֫օը֛֧ "(Who is shri. Pranab Mukharji ?)

332

Natural Language Processing

֏֞֒ֆ ׂ‫ ׆‬ճչ᭭ց ׂ‫֫֒ ׈ׅ׊‬վ֠ ᭭֗ֆեᮢ տ֞֔֞(India got freedom on 15th August 1947). OXWSXW: x շ֫օ ׂ‫ ׆‬ճչ᭭ց ׂ‫֫֒ ׈ׅ׊‬վ֠ ᭭֗ֆեᮢ տ֞֔֞ " Who got freedom on 15th August 1947? 2.

3. ֆ֞վ֐֛֞֔ըᮕ֑֧֞և֧ը֛֧ Tajmahal is at Agra. Output: x ֆ֞վ֐֛֞֔շ֫ւ֧ ը֛֧:KHUHLV7DMPDKO "

3. QUESTION GENERATION PROCESS To generate question automatically in Marathi language using rule based approach as used in structured abstractive text summarization. In rule based approach handcrafted rules are created according to grammatical rules of Marathi language to generate question from input sentence. The system generate the TXHVWLRQ VWDUWV ZLWK WKH ZRUGV ³շ֫ւ֧ :KHUH ´ ³շ֫օ ZKR ´ ᳰշֆ֠ (how)³շ֧ ᭪֛֞ :KHQ ´HWF7KHV\VWHPFDQQRWJHQHUDWHWKH question starts with ³շ֞ ZK\ ´³շ֚֞KRZ ´HWF By following these rules, for the given sentence, the system mainly tries to generate shallow questions with question words: ³շ֑֞ kay, :KDW ´(QJOLVK0HDQLQJ ³շ֧ ᭪֛֞ (kevha:KHQ ´ ³շ֫ւ֧ (kothe :KHUH ´ ³շ֫օ  շ֫օ֠ NRQNRQL :KR:KRP ´ ³ᳰշֆ֠(kiti+RZPXFK+RZPDQ\ ´HWF [3, 4].

4.

If any cardinal or ordinal or integer or the numbers in word and the word is neither preceded nor followed by the word which is in the list of months in input sentence, then ᳰշֆ֠ (kiti, How much/How any) type question generate from input sentence. For example: ͙00 ֗֙ᭅ (200YDUVK,͙00 years), 10 ᳰֈ֚֗ (͘0 GLYDV, 10 days), ͜ ćĭĝ (͜WDV , 5 hours), ֈ֫֊ ֔֫շ (GRQ ORN, two persons). 5. If question has been generated, then the punctuation marks ´³(֌֢օᭅ֟֗֒֞֐, VWRSIXOO) replace with ³?´ (ᮧ᳤֟ռ᭠֛Question mark) [2] HRw to generate question for input sentence using rule based system is shown in following flowchart:

Fig 1. Question Generation Process

4. RULE BASED APPROACH Rule based approach play important role in question generation system. It contains conditional statements which are used to generate the question from input text [4, 5, 8]. Rules are created to generate question from given Marathi text: 1. If the noun referring to person name (the preceding word may or may not in the prefix or designation and middle or last name) and whose preposition is ռ֞ (cha),ռ֠ (chi),ռ֧ (che), ֊֧ (ne), ᭒֑֞ (chya), ᭜֑֞(tya), etc., in the given sentence, then type շ֫օ շ֫օ֞ռ֞ շ֫օ֞ռ֠շ֫օ֞᭒֑֞ (kRQ, NRQDFKD NRQDFKL NRQDFK\D Who/Whom) type question generate from input sentence. For Example: փ֩ է᭣᭞ֈ֡֔ շ֔֞֐ ( Dr. Abaddul Kalam,  Dr. Abaddul Kalam) ֑֡֗֒֞վ ᳲ֚եչ֔֞ \XUDM VLQJK OD to RQOLQH@ $YDLODEOH https://cpl.usu.edu/htm/publications/publication=14766, 2003. [10] *LWHOVRQ$$0HU]O\DN01³6LJQDWXUHDQDO\VLVRIOHDI reflectance spectra: algorithm development for remote VHQVLQJ RI FKORURSK\OO´ - 3ODQW 3K\VLRO 9RO SS 494±500, 1996. [11] Champagne C., et al., "Mapping Crop Water Status: Issues of Scale in the Detection of Crop Water Stress Using Hyperspectral Indices", Proceedings of the 8th International Symposium on Physical Measurements and Signatures in Remote Sensing, Aussois, France, pp. 79-84, 2001.

This work is supported by Department and Science and Technology under the Funds for Infrastructure Technology (DST-IST) with sanction no. SR/FST/ETI-340/2013 to department of Computer Science and Information Technology. Dr. Babasaheb Ambedkar Marathwada University, Aurangabad, Maharashtra, India. The authors would like to thank Department and University Authorities for providing the infrastructure and necessary support for carrying out the research.

6. REFERENCES [1] Anatoly A. Gitelson, AndrHV 9LQ×D 9HURQLFD &LJDQGD 'RQDOG&5XQGTXLVWDQG7LPRWK\-$UNHEDXHU³5HPRWH HVWLPDWLRQ RI FKORURSK\OO FRQWHQW LQ SODQWV´ *HRSK\VLFDO Research Letters, Vol. 32, 2005. [2] 3UDMDNWD3DWDQH$QXS9LEKXWH³&KORURSK\OODQG1LWURJHQ Estimation Techniques: A 5HYLHZ´ ,QWHUQDWLRQDO -RXUQDO of Engineering Research and Reviews, Vol. 2, pp. 33-41, 2014. [3] C. Lin, S. C. Popescu, S. C. Huang, P. T. Chang, and H. /:HQ ³$ QRYHO UHIOHFWDQFH-based model for evaluating chlorophyll concentrations of fresh and water-stressed OHDYHV´%LRJHRVFLHQFHV9ROSS±66, 2015. [4] Zarco-Tejada, P.J., Miller, J.R., Mohammed, G.H., Noland, 7/ 6DPSVRQ 3+ ³2SWLFDO LQGLFHV DV ELRLQGLFDWRUV RI IRUHVW FRQGLWLRQ IURP K\SHUVSHFWUDO &$6, GDWD´ ,Q Proceedings of the 19th Symposium of the European Association of Remote Sensing Laboratories (EARSeL), Valladolid, Spain, 1999. [5] .LP06'DXJKWU\&67&KDSSHOOH(:³7KHXVHRI high spectral resolution bands for estimating absorbed SKRWRV\QWKHWLFDOO\ DFWLYH UDGLDWLRQ $SDU ´ ,n: Proceedings of the Sixth Symposium on Physical Measurements and Signatures in Remote Sensing, Val '¶,VHUH)UDQFHSS±306, 1994.

386

Remote Sensing and GIS, Smart City and Smart Villages

Chapter 79

Land Use Land Cover Mapping and Change Detection by Using Remote Sensing LISS-III Temporal Datasets of Aurangabad City Ajay D. Nagne

Rajesh K. Dhumal

Amol D. Vibhute

K. V. Kale

S. C. Mehrotra

Dept. of CS&IT, r.B.A.M.University, Aurangabad (MS), INDIA.

Dept. of CS&IT, Dr.B.A.M.University, Aurangabad (MS), INDIA.

Dept. of CS&IT, Dr.B.A.M.University, Aurangabad (MS), INDIA.

Dept. of CS&IT, Dr.B.A.M.University, Aurangabad (MS), INDIA.

Dept. of CS&IT, Dr.B.A.M.University, Aurangabad (MS), INDIA.

ajay.nagne@gmail. [email protected] amolvibhute2011@ com om gmail.com

kvkale91@gmail. com

mehrotra.suresh15j @gmail.com

ABSTRACT The objective of this paper is to present quick practical approach for Mapping and Analysis of Land Use Land Cover change detection in Aurangabad Municipal Corporation (AMC). The Remote Sensing and GIS technology are widely used in identification of LULC and monitoring of urban land use. For this purpose Temporal LISS-III Multispectral datasets of December 2003 and February 2015 were used. The area was categorized into six types. A Maximum Likelihood supervised classifier was used to classify both images. It has provided satisfactory results. The overall accuracy with the classifier was found to be 73% and 93% with Kappa Coefficient 0.64 and 0.90 for the year 2003 and 2015, respectively. In Residential area, it was found to be growth of 11.34% within 12 years of time spam. It may be because of deficient rainfall during the period. Because of this, water body area has been decreased by 1.89% in 2015. Less rainfall causes decrement in vegetation area, Fallow Land by 3.69% and 33.98% in 2015 respectively and increment in Barren Land by 24.94% in 2015.

General Terms Remote Sensing, Geographical Information System

Keywords Land Use Land Cover (LULC), Maximum Likelihood Classifier (MLC), LISS-III.

With the use of remote sensing and Geographical Information System (GIS) methods, land use/spread mapping has given a valuable and gritty approach to enhance the choice of territories intended to horticultural, urban and/or modern zones of an area. Use of remotely sensing information rolled out conceivable to consider the improvements in area spread in less time, requiring little to no effort and with better precision in relationship with GIS that gives appropriate stage to information examination, redesign and recovery. The images with high spatial resolution and advancement in digital image processing tools and GIS technology can identify a perfect/ actual LULC of the region [7] [8] [9]. The Section 2 has briefly mentioned information regarding the area studied and the remote sensing data used for this study. The section 3 provides detail about Methodology. The results regarding change analysis and accuracy are given in section 4 and 5. The paper is concluded in section 6.

2. STUDY AREA AND DATA SETS Aurangabad district (see Figure 1) is situated in the north central SDUWRI0DKDUDVKWUDEHWZHHQ1RUWK/DWLWXGHƒ¶DQGƒ DQG (DVW /RQJLWXGH ƒ ¶ DQG ƒ ¶ 7KH ZRUOG IDPRXV Ajanta and Ellora caves are situated in Aurangabad district. There are also a few caves near Aurangabad City. Other monuments of national fame are Bibi-ka-Maqbara and Daulatabad fort [10] [11].

1. INTRODUCTION Digital change detection techniques by using multi-temporal satellite imagery helps in understanding Land Use Land Cover dynamics. Land cover refers to the physical characteristics of HDUWK¶VVXUIDFHFDSWXUHGLQWKHGLVWULEXWLRQRIYHJHWDWLRQZDWHU soil, rock and other physical features of the land, including those created solely by human activities e.g., settlements. While land-use alludes to the path in which land has been utilized by people and their environment, more often than not with accent on the useful part of area for financial exercises [1] [2] [3]. The area use/spread example of a locale is a result of normal and financial components and their use by man in time and space. Data ashore utilize/spread and potential outcomes for their ideal use is key for the choice, arranging and usage of area use plans to meet the expanding requests for fundamental human needs and welfare. This data additionally helps with checking the elements of area use coming about out of changing requests of expanding populace [4] [5] [6].

Figure 1 Study Area

387

Cognitive Knowledge Engineering

The atmosphere of the Aurangabad district is a hot summer and a general dryness during the year. The winter season start from the end of November when temperatures starts to fall. December is the coldest month. The daily temperature increases continuously from the beginning of March and May is the hottest month. Considering an atmospheric condition of Aurangabad region a December, January and February are the best months to identify a Land Use Land Cover by using satellite images of this study area. The Temporal datasets of LISS-III December 2003 and February 2015 have been chosen and obtained from NRSC Hyderabad, India [12] [13] [14].

2.1 LISS-III Data Set The LISS-III (Linear Imaging Self Scanning Sensor) sensor is an optical sensor working in four spectral bands (green, red, near infrared and short wave infrared). LISS- III is a mediumresolution multispectral camera. It covers a 141km-wide swath with a resolution of 23.5 metres in all spectral bands. The detailed information of LISS-III sensor image is shown in the Table 1 [15]. Table 1 LISS-III sensor information Sr. No. 1.

Parameters

LISS-3

Swath

141 Km Band2 0.52 ± 0.59

2.

Spectral band (micron)

have a same spatial resolution. For present study a temporal data set of LISS-III December 2009 and LISS-III February 2015 were used. In the first step Layer stacking was performed because, Raw LISS-III Remote Sensing data has four bands and there is a separate file (.tiff) for every band. ENVI tool was used to perform a layer stacking. After getting a single tiff file a subset operation was performed for this purpose Aurangabad city boundary shape file was used. Now subset represents only a Aurangabad Municipal Corporation Area.

3.1 Classification The maximum likelihood supervised classification algorithm takes advantages of probability density functions which are used in the classification. It works on overlapping signatures with the help of probability. This classifier is based on Bays theorem, in which a pixel belongs to maximum likelihood, is categorized into the related class. The probability density functions are estimated by two weighting factors through Bayes theorem. Firstly, the user trains the a priori probability or specific signatures of the class in the given image. Secondly, for each class the cost of misclassification is weighted as per its probability. Both factors outcomes better decreasing the misclassification. This classifier classifies spectral response patterns of an unknown pixel through estimating both the variance and covariance of the class. The user must have the knowledge about the spectral signature or ground truth [16] [17] [18] [19].

Band3 0.62 ± 0.68 Band4 0.77 ± 0.86 Band5 1.55 ± 1.70

3. Methodology The Proposed methodology for Land Use Land Cover classification for identification of change detection is shown the Figure 2.

Figure 3 LISS-III of 2003 and 2015 classified image

4. CHANGE ANALYSIS

Figure 2 Methodology The required data sets were collected from the NRSC Hyderabad, India. For identification of change in LULC it required to have an at least two Temporal datasets and it should

Urban expansion has increased the exploitation of natural resources and has changed land use and land cover patterns. The process of change detection depends on the phenomenon or scene at different times. Change detection techniques are used to indicate the classes between the input images to detect the major changes that occurred in the study area. Land cover is a critical element in change studies, affecting many aspects of the environmental system. The point of the present study is to order the LULC as well as recognize the change recognition. Powerful use of RS for LULC change disclosure, as it were, depends on upon a legitimate comprehension of the study area. Table 2 shows a Change detection of Temporal LISS-III datasets of December 2003 and February 2015, it shows all results in percentage and also shows changes in each class. In Residential area there is a growth of 11.34% is identified within 12 years of time spam. Since From decades every year the less Rainfall by some percentage was identified till present (i.e. 2015) in Aurangabad Region. Because of less rainfall it is clearly observed that, water body area has been decreased by 1.89% in 2015. Less rainfall also causes decrement in vegetation area and Fallow Land by 3.69% and 33.98% in 2015

388

Remote Sensing and GIS, Smart City and Smart Villages

respectively. Due to decrement in Vegetation and Fallow Land Barren Land was increased by 24.94% in 2015. Table 2 Change Detection of LISS-III Jan 2009 and Feb 2015 LISS-III LISS-III FEB DEC 2003 2015 Classificati Color Classification Difference on % of Each % of Each Classes Classes

Class Name

Residential

Magenta

28.50

17.16

11.34

Rock (hill without Vegetation)

Cyan

21.62

18.35

3.28

Water Body

Blue

0.30

2.20

-1.89

Vegetation

Green

8.89

12.58

-3.69

Barren Land Yellow

35.15

10.21

24.94

5.54

39.52

-33.98

Fallow Land

Red

5. ACCURACY ASSESSMENT The error matrix-based accuracy assessment method is the most common and valuable method for the evaluation of change detection results. Thus, an error matrix and a Kappa analysis were used to assess change accuracy. Accuracy assessment is comparison of a classification with ground truth data to evaluate how well the classification represents the ground data. For LULC change detection analysis an accuracy assessment is very important to understand and calculate the change accurately. For this purpose it required a proper number of ground truth (samples) for every class [18] [19] [20]. In this accuracy assessment ten samples (ground truth) for every class were. Accuracy assessment was performed on Maximum Likelihood classifier and it has given a good result in both the data sets. In January 2015 classification a Maximum Likelihood classier has given a accuracy of 93.43% with Kappa coefficient 0.90, whereas in 2003 classification Maximum Likelihood classifier has given accuracy of 73.07% with Kappa Coefficient 0.64. Table 3 shows an Overall Accuracy and corresponding Kappa Coefficient. Table 3 Overall Accuracy and Kappa Coefficient

Classifier Name Maximum Likelihood

LISS-III January 2009 LISS-III February 2015 Overall Accuracy Overall Accuracy Percent

Kappa Coefficient

Percent

Kappa Coefficient

73.07

0.64

93.43

0.90

6. CONCLUSION Satellite Remote sensing systems have an ability to cover a large area repetitively. The point of the present study is to order the LULC as well as recognize the change in the Aurangabad Municipal Corporation area. For classification an objects are categorized in to six types. A Maximum Likelihood supervised classifier has given a good overall accuracy of 73% and 93% with Kappa Coefficient 0.64 and 0.90 for the year 2003 and 2015, respectively. In 12 years of time spam an 11.34% growth was identified in Residential area. From past decade the less

Rainfall was noted by every year by some percentage in Aurangabad Region. Because of this less rainfall, water body, Vegetation and Fallow Land was decreased by 1.89%, 3.69% and 33.98% in 2015 respectively. Due to decrement in Vegetation and Fallow Land Barren Land was increased by 24.94% in 2015. Thus it can be concluded that Remote Sensing technology can be used effectively for monitoring temporal and spatial changes in LULC.

7. ACKNOWLEDGMENTS Authors would like to acknowledge 1)UGC - BSR Fellowships 2)DST_FIST and 3)UGC-SAP(II)DRS Phase-I and Phase- II F.No.-3- 42/ 2009 and 4- 15/ 2015/DRS- II for Laboratory Facility to Department of CS&IT, Dr. B.A.M. University, Aurangabad(MS),INDIA.

8. REFERENCES [1] Bhagawat Rimal, Application Of Remote Sensing And GIS, Land Use/Land Cover Change In Kathmandu Metropolitan City, Nepal, Journal Of Theoretical And Applied Information Technology, 2011. [2] A.A. Belal and F.S. Moghanm, Detecting urban growth using remote sensing and GIS techniques in Al Gharbiya governorate, Egypt, The Egyptian Journal of Remote Sensing and Space Sciences (2011) 14, 73±79. [3] O.S. Olokeogun, O.F. Iyiola And K. Iyiola, Application Of Remote Sensing And GIS In Land Use/Land Cover Mapping And Change Detection In Shasha Forest Reserve, Nigeria, The International Archives Of The Photogrammetry, Remote Sensing And Spatial Information Sciences, Volume Xl-8, 2014 ISPRS Technical Commission Viii Symposium, 09 ± 12 December 2014, Hyderabad, India. [4] Daniel Ayalew Mengistu and Ayobami T. Salami, Application of remote sensing and GIS inland use/land cover mapping and change detection in a part of south western Nigeria, African Journal of Environmental Science and Technology Vol. 1 (5), pp. 099-109, December, 2007. [5] Rawat, J. S., and Manish Kumar. "Monitoring land use/cover change using remote sensing and GIS techniques: A case study of Hawalbagh block, district Almora, Uttarakhand, India." The Egyptian Journal of Remote Sensing and Space Science 18.1 (2015): 77-84. [6] Nayana S. Ratnaparkhi, Ajay D. Nagne and Bharti Gawali, Analysis of Land Use/Land Cover Changes Using Remote Sensing and GIS Techniques in Parbhani City, Maharashtra, India, International Journal of Advanced Remote Sensing and GIS 2016, Volume 5, Issue 4, pp. 1702-1708, Article ID Tech-573 ISSN 2320 ± 0243. [7] Nayana S. Ratnaparkhi, Ajay D. Nagne, Dr. Bharti Gawali, A Land Use Land Cover classification System Using Remote Sensing data, International Journal of Scientific & Engineering Research, Volume 5, Issue 7, July-2014. [8] Hegazy, Ibrahim Rizk, and Mosbeh Rashed Kaloop. "Monitoring urban growth and land use change detection with GIS and remote sensing techniques in Daqahlia governorate Egypt." International Journal of Sustainable Built Environment 4.1 (2015): 117-124. [9] Ajay D. Nagne and Dr. Bharti W. Gawali, Transportation Network Analysis by Using Remote Sensing and GIS a Review, International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 3, Issue 3, May-Jun 2013, pp.070-076.

389

Cognitive Knowledge Engineering

[10] Kashid, Sumedh D., Ajay D. Nagne, And K. V. Kale. "Solid Waste Management: Bin Allocation and Relocation by Using Remote Sensing & Geographic Information System.", International Journal of Research in Engineering and Technology, Volume: 04 Issue: 12 | Dec-2015.

And Gis, International Journal of Engineering and Management Sciences, Vol.3 (4) 2012: 513-519.

[11] Dhananjay B. Nalawade, Sumedh D. Kashid, Rajesh K. Dhumal, Ajay D. Nagne, Karbhari V. Kale, Analysis of Present Transport System of Aurangabad City Using Geographic Information System, International Journal of Computer Sciences and Engineering Vol.-3(6), PP(124128) June 2015, E-ISSN: 2347-2693. [12] U.S. Balpande, Ground Water Information Aurangabad District Maharashtra, Government of India Ministry of Water Resources Central Ground Water Board, 1791/DBR/201, http://cgwb.gov.in/District_Profile/Maharashtra/Aurangaba d.pdf. [13] Ajay D. Nagne, Rajesh K. Dhumal, Amol D. Vibhute, Yogesh D. Rajendra, K. V. Kale, S. C. Mehrotra, Suitable Sites Identification for Solid Waste Dumping Using RS and GIS Approach: A Case Study of Aurangabad, (MS) India, The 11th IEEE INDICON 2014 Conference, Pune Dec 2014. 978-1-4799-5364-6/14/$31.00, 2014 IEEE. [14] Ajay D. Nagne, Amol D. Vibhute, Bharti W. Gawali, Suresh S. Mehrotra, Spatial Analysis of Transportation Network for Town Planning of Aurangabad City by using Geographic Information System, International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July 2013. [15] Resourcesat ±  'DWD 8VHU¶V +DQGERRN 1DWLRQDO 5HPRWH Sensing Agency, Department Of Space, Govt. Of India Nrsa Balanagar, Hyderabad - 500037, A.P. India. Http://Bhuvan.Nrsc.Gov.In/Bhuvan/Pdf/Resourcesat1_Handbook.Pdf. [16] Amol D. Vibhute, Rajesh K. Dhumal, Ajay D. Nagne, Yogesh D. Rajendra, K. V. Kale and S. C. Mehrotra, ³$QDO\VLV &ODVVLILFDWLRQ DQG (VWLPDWLRQ RI 3DWWHUQ IRU Land of Aurangabad Region Using High-Resolution 6DWHOOLWH ,PDJH´ 3URFHHGLQJV RI WKH 6HFRQG ,QWHUQDWLRQDO Conference on Computer and Communication Technologies, Advances in Intelligent Systems and Computing AISC Series 11156, Vol. 380, pp 413-427, Springer India. DOI 10.1007/978-81-322-2523-2_40, 04 September 2015. [17] Amol D. Vibhute, Ajay D. Nagne, Bharti W. Gawali, 6XUHVK & 0HKURWUD ³&RPSDUDWLYH DQDO\VLV RI GLIIHUHQW supervised classification techniques for spatial land XVHODQG FRYHU SDWWHUQ PDSSLQJ XVLQJ 56 DQG *,6´ International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013. [18] Foody, Giles M. "Status of land cover classification accuracy assessment." Remote sensing of environment 80.1 (2002): 185-201. [19] Rajesh K. Dhumal, Amol D. Vibhute, Ajay D. Nagne, Yogesh D. Rajendra, Karbhari V. Kale and Suresh C. Mehrotra, Advances in Classification of Crops using Remote Sensing Data, Cloud Publications International Journal of Advanced Remote Sensing and GIS 2015, Volume 4, Issue 1, pp. 1410-1418, Article ID Tech-483 ISSN 2320 ± 0243. [20] Ezeomedo, I. C. And Igbokwe, J. I., Mapping And Analysis Of Land Use And Land Cover For A Sustainable Development Using Medium Resolution Satellite Images

390

Remote Sensing and GIS, Smart City and Smart Villages

Chapter 80

Review of Digital Soil Mapping Procedures Kayte Jaypalsing Natthsuing

R. R. Deshmukh

Department of CS and IT, Dr. B. A. M University, Aurangabad- India.

Department of CS and IT, Dr. B. A. M University, Aurangabad- India.

[email protected] m

rrdeshmukh.csit@ bamu.ac.in

ABSTRACT Digital soil mapping is a soft way to finding all information of soil contains.This paper we are review of hardware, software, and methodology, for digital soil mapping, In the ¶VDERYH 65 percent of the population in India working as comprised of cultivators(farmers) and agricultural laborsignifying the FRXQWU\¶V GHSHQGHQFH RQ DJULFXOWXUH 7KHUHIRUH WKH QHHG IRU soil information using remote sensing resources for development over all social and economic development was duly recognized from the very beginning. Today all farmers want information about our soil, which types of contain are available, which crop are good for land so many question by former, so we are try to finding the solution using digital soil mapping techniques.

Keywords:

Digital Soil Mapping (DSM), Geographical Information System (GIS), Remote Sensing (RS), Principal Component Analysis (PCA)

1

INTRODUCTION

Soil is a bunch of mineral contains, we need to explore information of soil need traditional manual process, but digital soil mapping is a process to give more information of soil without physical contact using remote sensing data (example of ASTER database, Landsat-7 database). Digital Soil Mapping is a process for solving soil problem and got the much information about. There is a global need for quantitative soil information for environmental monitoring and modeling. One response to this demand is digital soil mapping, where soil mapping are produced digital based on environmental variables[1] the environmental or scorpan factors[1] derived from various sources (Digital elevation models, remote sensing images, existing soil maps), and available in digital form, are used to generate soil information in the form of a database where most of the information consists of prediction that are statistically optimal. Soil mapping in general requires (i)a predefined model of soil formation,(ii) data on soil properties and on other environmental variables that have significant impact on soil formation and thus on the spatial distribution of the soil properties.Digital Soil Mapping ± is the computer-assisted production of digital maps of soil type and soil properties. It typically implies use of mathematical and statistical models that combine information from soil observations with information contained in correlated environmental variables and remote sensing images[2], [1]. In this sense, traditional soil mapping and digital soil mapping do not differ much. Both approaches need input data on soil and covariates characterizing the environment where the soil

Stephen G. Karungaru Institute of Advanced Science & Technology, The University of Tokushima, Tokushima, Japan.

N. V. Kalyankar Gondwana University Gadchiroli

drkalyankarnv@yah oo.com

[email protected] a-u.ac.jp formation takes place. The major difference is the way how the model derives the soil information from the input data. The traditional models are based on empirical studies and qualitatively defined correlation that formulates a rational model LQWKHVXUYH\RU¶V PLQGXVHGWRXQGHUVWDQGDQGFKDUDFWHUL]HWKH soil resources. Decisions are made mainly on the field, where all environmental covariates can be directly observed and information on the soil can be deduced. The digital soil mapping approach is quite similar; it is based on hard soil data as well. Like in the traditional approach, profile information is needed to train our models, and to understand the soil resources of the area. The major differences, the strengths and also the limitation are coming from the way how the environmental covariates are represented in the procedure. Digital Soil Mapping requires digital data source as input variables for the quantitative models. -HQQ\¶Vwell ±known equation (1941) identified 5 major factors in the soil formation, namely the climate, organism, relief, parent material and time: S=f(cl,o,r,p,t) The prediction of the soil variables and a successful survey needs good quality, adequate resolutiRQ LQSXW GDWD -HQQ\¶V approach focused on the predication of certain soil chemical, physical or biological characteristics on a given location and did not consider the soil as a continuum, where the soil properties at a given location depend on their geographic position and also on the soil properties at neighboring locations. This fact is utilized by geo-statisticians, who predict soil properties of a given site for known observations neighboring the point. From an applied soil survey point of view, the group of the five soil forming factors needs to be enlarged with the addition of the geographic position. Some soil properties are difficult or expensive to measure, but can be predicted with acceptable accuracy from other soil parameters of the same location. That we also have to consider, where a full picture has to be painted about the data needs for soil property. This approach was followed and summarized by[1], he identified 7 factors for soil spatial predication and formulated the so called SCORPAN equation:

Sc =f(s,c,o,r,p,a,n) or Sa = f(s,c,o,r,p,a,n) Where Sc is soil classes and Sa is soil attributes. The S refers to soil information either from a prior map, or from remote or proximal sensing or expert Knowledge.

391

Cognitive Knowledge Engineering

1.2 Soil Important

1.5 Use of soil information

Soil is amixture of organic matter, minerals, liquids, gases, and countless organisms that together support life on Earth.The importance of soil can be explained in terms of the following three main components:

Soils are a primary source and, as they are directly or indirectly linked to the existence and development of society, soils are back on the global agenda [6]A thorough knowledge of soils and their formation and distribution in the landscape is of the utmost importance to maintain ecosystem service quality and protect soil itself from the above-mentioned threats.We are review some point in next section of literature review how work digital soil mapping.

1.3 Ecosystem services/soil functions Soil provides ecosystem services to sustain life on earth. It provides food, fiber and a place to live, and is also the foundation for all terrestrial transformation and fluxes[3][4]. listed six major ecosystem services supplied by the soil, which are: (1) buffering and moderation of the hydrological cycle; (2) physical support of plants; (3) retention and delivery of nutrients to plants; (4) disposal of wastes and dead organic matter; (5) renewal of soil fertility; and (6) regulation of major element cycles.

2

REVIEW OF HARDWARE

In this paper we review various technologies, methodology, hardware, software for digital soil mapping. By hardware we mean various kinds of sensor and instrument which can give better soil and scorpan data, and software we mean mathematical or statistical models which can improve spatial predictions. Start followed by discursion on the implication of using data-mining tools for the production of digital soil maps.

2.1 Data Sources for Scorpan This data sources Scorpan are upgraded by [7]to [1], The scorpan factors can be obtained from various sensors, either remotely or proximally sensed. Remote sensing for soil properties is reviewed by[8], while proximal sensing is given by[9]. Here we list some sensors, based on their platform, that are commonly used for digital soil mapping:

2.1.1

Satellite based System

2.1.1.1 Hyperion 8.1.1 Hyperion the Hyperion from EO-1 satellite provides a high resolution hyperspectral imager capable of resolving 220 spectral bands from 400 to 2500nm with a 30m spatial resolution, and image swath width 7.5km. Hyperspectral images measure reflected radiation at a series of narrow and contiguous wavelength bands its use for digital soil mappingis still limited[10] and can be challenging as noise of the spectra and the influence of vegetation.

Fig. 1 Linkages of soil to different aspects of life[5]. This Figure 1 are highlights the importance and interaction of soils with aspects of life ranging from food production to the state where soil also acts as a gene pool, The seven functionsthat the soil performs are: biomass production, carbon pool, a source of raw materials, biodiversity pool, natural heritage, physical and cultural habitat, and storing and filtering of nutrients and water. These functions need to be preserved to ensure the sustained supply of ecosystem services from soils, so soil are important for digital soil mapping.

1.4 Soil threats A meanness of the ecosystem services they provide, soils are being constantly threatened by a number of natural or anthropogenic activities causing severe damage to soils, leading to loss of its capacity to provide the services in an effective way. This communication lists eight main threats to soils: erosion, decline in organic matter, soilcontamination, soil sealing, soil compaction, decline in soil biodiversity, salinesation, and floods and landslides.

Figure 2 satellite based remote sensing instruments as a function of wavelengths. The gray curve represents atmospheric electromagnetic opacity

2.1.1.2 ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) 8.1.2 The ASTERis a multispectral imaging system[11]. Multispectral Imagers measure radiation reflected from a

392

Remote Sensing and GIS, Smart City and Smart Villages

surface at a few wide, separated wavelength bands. ASTER measures visible reflected radiation in three spectral bands (VNIR between 0.52 and 0.86µm, with 15-m spatial resolution), and infrared reflected radiation in six spectral bands (SWIR between 1.6 and 2.43µm, with 30-m spatial resolution). In addition, ASTER records the data in band 3B (0.76±0.86µm) with a backward looking that enables the calculation of digital elevation model (DEM). ASTER also receives emitted radiation in five spectral bands (TIR between 8.125 and 11.65µm, with 90-m spatial resolution). It has been used for mapping geological units[12], and areas of degraded land.

2.1.1.2 Landsat TM, 8.1.3 The Landsat TMEnhanced Thematic Mapper Plus (ETM+)The Enhanced Thematic Mapper Plus (ETM+) is a multispectral scanning radiometer that is carried on board the Landsat 7 satellite. It provides images with spatial resolution of 30m for the visible and near-infrared, and 60m for the thermal infrared, and 15m for the panchromatic. Landsat has been used most often in digital soil mapping[13].illustrates its application for land-use mapping.

2.1.1.3 SPOT 8.1.4 6DWHOOLWHV 3RXU O¶ 2EVHUYDWLRQ GH OD 7HUUH RU (DUWK-

AVIRISis an airborne imaging instrument producing 224 spectral bands ranging from 400 to 2500 nm, with a spatial resolution of 20m. These AVIRIS spectra can be used to discriminate between soil types. 2.2.3 Airborne gamma radiometric Airborne gamma radiometricis Variations of gamma radiation has been found to correspond with the distribution of soilforming materials over the landscape. 2.2.4 Aerial photography Aerial photographythisis technique, providing images in the visual light, is still being used in soil surveys and with interpretation is used to generate soil maps.

2.3 Proximal, ground-based System 2.2.5 Electrical magnetic induction (EMI) EMI These instruments measure the bulk soil electrical conductivity, it has been successful for high resolution digital soil mapping for properties such as clay and water content[19].

2.3.1 Gamma radiometrics rayspectrometers

Gamma-

observing Satellites (SPOT), SPOT provides high-resolution multispectral images with resolution of 10m in the visible and near infra-red (0.50±0.89 µm), and 20m in the short wave infrared (1.58±1.75µm[14].Investigated its use for mappingsoil texture class.

Gamma Radiometricscan measure an energy spectrum ranging from 0 to 3 MeV. The value of gamma-ray spectrometry lies due to the different rock types contain varying amounts of radioisotopes of K, U and Th. Ground-based gamma ray spectrometers have been used for mapping soil properties.

2.1.1.4 AVHRR(Advanced

3

Very High Resolution Radiometer) 8.1.5 The AVHRR provides four to six bands of multispectral images (visible red, near infra-red, short-wave infra-red, and thermal infra-red) with 1.1 km resolution from the NOAA polar-orbiting satellite series. The AVHRR data have been collected to monitor global change information, however the data can be used as a cost-effective way of estimating soil properties at regional level[15].

2.1.1.5 MODI 8.1.6 Moderate

Resolution Imaging Spectroradiometer, MODIS is an instrument aboard the Terra and Aqua satellites. 7HUUD¶VRUELWDURXQGWKH(DUWKSDVVHVIURPQRUWKWRVRXWKDcross the equator in the morning, while Aqua passes south to north over the equator in the afternoon. Terra MODIS and Aqua 02',6DUHYLHZLQJWKHHQWLUH(DUWK¶VVXUIDFH HYHU\±2 days, acquiring data in 36 spectral bands at a resolution of 250m (620±876 nm), 500m (459±2155 nm), and 1000m (405± 14385nm). Its use mainly for monitoring vegetation activity via NDVI[16].[17]used MODIS data to derive surface albedo for the arid areas of Northern Africa and the Arabian peninsula.

2.2 Airborne based System 2.2.1 HyMapTM(HyperspectralMapping) TM HyMap is an airborne imaging VNIR-SWIR spectrometer, with 450±2500nm spectral coverage, 128 spectral bands of 10± 20nm bandwidths. Examples of its use for mapping soil are [18]. 2.2.2

AVIRIS (Airborne Visible Infrared Imaging Spectrometer)

REVIEW OF LITERATURE

Digital soil mapping is a process of upgrading information about soil with physical, laboratory, and remote sensing methodologies, of howthose methodologies are used in various papers we are review hear. 3.1 Brendan Malone et al[20], This paper are show the practical device for digital soil morphometric involves the application of new tools and techniques for the routine charactersation of soil profiles[21]. These tools and techniques include soil proximal sensing and spectroscopic instruments, and the ensuing data processing PHWKRGVQHHGHGWRXOWLPDWHO\JHQHUDWHXVHIXOµVRLO¶LQIRUPDWLRQ The portability of instruments such as vis-NIR and portableXRF spectrometers makes them natural candidates for routine Soil profile characterization. [20]Arebuilding a framework for making vis-NIR spectroscopy applicable for that purpose. This framework is embodied within a soil spectral inference engine with prototype tested upon two contrasting soil profiles at ground level field application of vis-NIR from enterprise (Nowley) in Australia. 3.2 Alex McBratney[1] This paper are indicates the following major soil properties showing relatively high correlation with remote sensing images: iron-oxide content, soil organic matter content, salt content, parent material differences, soil moisture content, and some chemical and physical properties like pH, calcium-carbonate, mineral N, total carbon are available in phosphorus, clay- siltand sand contents. Some soil properties are directly related to the surface colour and thus relatively easy to map when the soil is bare and visible spectra is used to detect the colour. Ironoxide and organic matter content, and partly the soil moisture contents and soil texture are good examples of that. Other soil features, like many of the chemical properties of the deeper horizons, can be detected only indirectly, through the type and

393

Cognitive Knowledge Engineering

the condition of the surface vegetation. These relationships are often indirect and explain less of the total spatial variation than the one for the soil surface properties. 3.3 F. Carré[22] F. Carreare clarify a schema for digital soil mapping, digital soil assessment and digital risk assessment. The schema shows a logical flow of data from the soil sampling to the end-user with appropriate feedbacks. As point wise 1] the combination of DSM with DSA is very powerful since DSA provides the basic requirements of DSM. DSM provides the necessary inputs to DSA. 2] The digital soil mapping community needs to be much more aware of end-user requirements and tailor their products accordingly. 3] The key features of these three interlinked new approaches are the carrying through and estimation of uncertainty at all stages of DSM, DSA and DSRA. 4] The digital soil mapping community must be prepared to educate end-users in the use of information and its associated uncertainty. [5] DSA and DSRA are not restricted to pedocentric goals of maintenance of soil quality but can be applied to many systems in which soil plays a part. 3.4 R. Taghizadeh-Mehrjardi[23] ThisPaper areusing decision tree technique. In this method, they have used a variety of environmental variables driven from different sources such as DEM, remote sensing and existing maps at Iran. 3.5 S.Y. Hong et al.[24] This paper study of clay content for soil organic carbon, soil were estimated based on the equal area smoothing spline and mapped corresponding soil series to produce spatial and depthwise soil property maps. As author view clay content showed better agreement in compared with recently sampled soils data for used methods soil organic carbon was measured by Tyurin method and clay content by pipette method with sodium hexameta phosphate as dispersant. These methods are measured and predicated top soil clay content showed good agreement with consistence improves for digital soil mapping. [25]

3.6 J.Padarian, B. Minasny et al This paper isevaluating methodology for global soil mapping; digital soil mapping has successfully being applied at a regional or national extent. In 2010 Minasny and McBratney[1] proposed a methodology for global soil mapping that is based on the data availability. While it has been used as a framework for DSM in large areas, there are further considerations when we want to map at a continental scale. This paper will discuss several considerations (1) soil data and covariates coverage, (2) spatial incompatibility of Scorpan layers, (3) extrapolation, (4) forms of the empirical functions, (5) uncertainty estimates, and (6) map validation. This paper isproviding a more detailed analysis of the use of scorpan-kriging approach for mapping soil properties in large areas. The general steps in scorpan-krigingmodeling involve: collection of a dataset of soil observations over the chosen area of interest; compilation of relevant covariates for the area; calibration or training of a spatial prediction function based on the observed dataset; interpolation and/or extrapolation of the prediction function over the whole area of interest; and finally validation using existing or independent datasets. The use of a scorpan-kriging approach to map large areas is feasible but it is necessary to have special considerations in some stages of the process. The amount of data required to model at this scale is considerable and it is usually necessary to use information from

different sources, harmonization.

with

the consequent

effort

of data

3.7 Alex. McBratney et al[1] Alex McBratneyareintroduce Scorpankriging method, This method is used for interpolation and limited extrapolation of VSDWLDOVRLOSRLQWGDWD7KHDVVXPSWLRQLVWKDWWKHVSDWLDO³WUHQG´ can be described by f(s, c, o, r, p, a, n) and the residuals e modelled byvariograms and a form of kriging. The final prediction is the sum of f() and e. Scorpan or regression kriging allows incorporation of both deterministic and stochastic components in kriging. These are based on methods that have been successfully applied in large areas.

4 CONCLUSION Digital soil mapping is a process to acquired soil information using remote sensing techniques and methodology that is useful for former and agriculture sectors. As according to HANS JENNY 1941[7] was generating the equation for finding soil information in single format but in 2003 Alex McBratney[1] are introduce f(scorpan)equation and as this equation are more possibility to study of soil using remote sensing. Overall this paper review of varies Remote Sensing Tools and Methodologies for study of Digital Soil Mapping using Remote Sensing. As our view ASTER satellite imaginary data and geological mapping have been studied in using only visible and near-infrared (VNIR) and short wave infrared (SWIR). ASTER Data for the Digital Soil Mapping are useful as compare to other satellite data. Technique Cal view PCA, is to extract the pertinent information from the different bands. This method removes the redundancy of information that exists between the different bands.

5 REFERENCES [1] A. B. McBratney, M. L. Mendon??a Santos, and B. Minasny, On digital soil mapping, vol. 117, no. 1±2. 2003. [2] ('RERV7+HQJODQG+5HXWHU³Digital soil mapping DVDVXSSRUWWRSURGXFWLRQRIIXQFWLRQDOPDSV´2IILFH IRU official publications of the European Communities, p. 68, 2006. [3] S. Naeem, F. S. C. Iii, R. Costanza, P. R. Ehrlich, F. B. Golley, D. U. Hooper, J. H. Lawton, R. V. O. Neill, H. a 0RRQH\ 2 ( 6DOD $ - 6\PVWDG DQG ' 7LOPDQ ³, VVXHVLQ(FRORJ\´,VVXHVLQ(FRORJ\YROQRSS± 12, 1999. [4] P. Daily, Gretchen. C., Alexander, Susan., Ehrlich, Paul. R., Goulder, Larry.; Lubchenco, Jane., Matson, Pamela. A., Mooney, Harold. A. and W. G. M. Sandra., Schneider, 6WHSKHQ+7LOPDQ'DYLG³(FRV\VWHP6HUYLFHV%HQHILWV 6XSSOLHG WR +XPDQ 6RFLHWLHV E\ 1DWXUDO (FRV\VWHPV´ Issues in Ecology, vol. 4, no. 4, pp. 1±12, 1999. [5] &2PXWR)1DFKWHUJDHOHDQG55RMDV³6WDWHRIWKH$UW Report on Global and Regional Soil Information: Where DUHZH":KHUHWRJR"´S [6] $ ( +DUWHPLQN ³6RLOV DUH EDFN RQ WKH JOREDO DJHQGD´ Soil Use and Management, vol. 24, no. 4, pp. 327±330, 2008. [7] H. Jenny, Factors of soil formation. A system of quantitative pedology, vol. 68, no. 4. 1994.

394

Remote Sensing and GIS, Smart City and Smart Villages

[8] E. Ben-'RU ³4XDQWLWDWLYH UHPRWH VHQVLQJ RI VRLO SURSHUWLHV´$GYDQFHVLQ$JURQRP\YROQR-XO\SS 173±243, 2002.

mapping of soil classes using decision trees in central ,UDQ´ 'LJLWDO 6RLO $VVHVVPHQWV DQG %H\RQGQR -XO\ SS 197±202, 2012.

[9] V. I. Adamchuk, J. W. Hummel, M. T. Morgan, and S. K. 8SDGK\D\D ³2Q-the-go soil sensors for precision DJULFXOWXUH´ &RPSXWHUV DQG (OHFWURQLFV LQ $JULFXOWXUH vol. 44, no. 1, pp. 71±91, 2004.

[24] S. Y. Hong, Y. H. Kim, K. H. Han, B. K. Hyun, Y. S. =KDQJ DQG . & 6RQJ ³'LJLWDO VRLO PDSSLQJ RI VRLO prRSHUWLHVIRU.RUHDQVRLOV´'LJLWDO6RLO$VVHVVPHQWVDQG Beyond, no. 1999, pp. 435±438, 2012.

[10] B. Datt, T. R. McVicar, T. G. Van Niel, D. L. B. Jupp, and - 6 3HDUOPDQ ³3UHSURFHVVLQJ (2-1 Hyperion hyperspectral data to support the application of agricultural LQGH[HV´ ,((( 7UDQVDFWLRQV RQ *HRVFLHQFH DQG 5HPRWH Sensing, vol. 41, no. 6 PART I, pp. 1246±1259, 2003.

[25] $%D%%$%0F%UDWQH\0F%UDWQH\³'LJLWDO6RLO 0DSSLQJ´ *HRGHUPD YRO  QR -DQXDU\ SS ±14, 2016.

[11] Y. Yamaguchi, A. B. Kahle, H. Tsu, T. Kawakami, and M. 3QLHO ³2YHUYLHZ RI DGYDQFHG VSDFHERUQH WKHUPDO emission and reflection radiometer $67(5 ´ ,((( Transactions on Geoscience and Remote Sensing, vol. 36, no. 4, pp. 1062±1071, 1998. [12] C. Gomez, C. Delacourt, P. Allemand, P. Ledru, and R. :DFNHUOH ³8VLQJ $67(5 UHPRWH VHQVLQJ GDWD VHW IRU JHRORJLFDO PDSSLQJ LQ 1DPLELD´ 3K\VLFV DQG &KHPLVtry of the Earth, Parts A/B/C, vol. 30, pp. 97±108, 2005. [13] A. E. Hartemink, A. McBratney, and M. MendoncaSantos, Digital Soil Mapping with Limited Data. 2008. [14] 0 * % (0%DUQHV ³0XOWLVSHFWUDO 'DWD )RU 0DSSLQJ 6RLO 7H[WXUH 3RVVLEOLWLHV DQG /LPLWDWLRQV´ Applied Engineering in Agriculture, vol. 16, no. 1, pp. 731±741., 2000. [15] , 2 $ 2GHK DQG $ % 0F%UDWQH\ ³8VLQJ $9+55 images for spatial prediction of clay content in the lower 1DPRL9DOOH\RIHDVWHUQ$XVWUDOLD´*HRGHUPDYROQR 3±4, pp. 237±254, 2000. [16] $ +XHWH & -XVWLFH DQG + /LX ³'HYHORSPHQW RI vegetation and soil indices for MODIS-(26´ 5HPRWH Sensing of Environment, vol. 49, no. 3, pp. 224±234, 1994. [17] ( D 7VYHWVLQVND\D ³5HODWLQJ 02',6-derived surface albedo to soils and rock types over Northern Africa and the $UDELDQSHQLQVXOD´*HRSK\VLFDO5HVHDUFK/HWWHUVYRO no. 9, pp. 3±6, 2002. [18] 7 6HOLJH - %|KQHU DQG 8 6FKPLGKDOWHU ³+LJK resolution topsoil mapping using hyperspectral image and field data in multivariate regression modelinJSURFHGXUHV´ vol. 136. pp. 235±244, 2006. [19] [19] ' / &RUZLQ DQG 6 0 /HVFK ³$SSDUHQW VRLO HOHFWULFDO FRQGXFWLYLW\ PHDVXUHPHQWV LQ DJULFXOWXUH´ Computers and Electronics in Agriculture, vol. 46, no. 1±3 SPEC. ISS., pp. 11±43, 2005. [20] B. Malone, A. Mcbratney, B. Minasny, and W. D. S. 0HQGHV ³$ VRLO VSHFWUDO LQIHUHQFH HQJLQH IRU GLJLWDO PRUSKRPHWULFFKDUDFWHULVDWLRQRIVRLOVௗ´YROQRS 2014, 2014. [21] $ ( +DUWHPLQN DQG % 0LQDVQ\ ³7RZDUGV GLJLWDO VRLO PRUSKRPHWULFV´ *HRGHUPD YRO ±231, pp. 305±317, 2014. [22] F. Carr??, A. B. McBratney, T. Mayr, and L. Montanarella, ³'LJLWDO VRLO DVVHVVPHQWV %H\RQG '60´ *HRGHUPD YRO 142, no. 1±2, pp. 69±79, 2007. [23] R. Taghizadeh-Mehrjardi, B. Minasny, a B. Mcbratney, J. Triantafilis, F. Sarmadian, and N. ToomaQLDQ³'LJLWDOVRLO

395

Cognitive Knowledge Engineering

Chapter 81

Prediction of Arsenic Content in Soil using Reflectance Spectroscopy Vaibhav N. Lokhande

Rahul T. Naharkar

Ratnadeep R. Deshmukh

Dept. of CS & IT Dr. BAMU,Aurangabad (MS), India

Dept. of CS & IT Dr. BAMU,Aurangabad (MS) India

Dept.of CS & IT Dr. BAMU,Aurangabad (MS) India

vaibhavlokhande04@gmail

[email protected]

[email protected] .in

.com ABSTRACT Contamination of soil is one of the serious problems in the recent days. It causes various effects on damage of environment and Health. It is necessary to obtain soil quality with metals. The research aims to predict content of arsenic in soil with the use of Fieldspec 4 Spectroradiometer. The instrument is used for obtaining spectral signature measurement of soil samples gathered from Aurangabad district of Maharashtra, India. The regression method Partial Least Squares regression is used for in the visible near infrared range to predict arsenic content in the soil samples. It is implemented with different spectral processing techniques which are Median Filter, Gaussian Filter, Savitzky-Golay smoothing, first derivative are used. The implementation provides best method for soil spectroscopy in VNIR range for prediction of arsenic (As) content in agriculture soil. The Normalized difference spectral index gives content value of arsenic in each soil samples.

The total soil samples are gathered from Aurangabad District of Maharashtra, India. The city located in Maharashtra state in western India. The city location is at 19° 53' N and 75° 23' E. The annual mean temperature at the study area ranges from 17 to 33 °C. Total 15 samples were collected from different agriculture lands in the district. They are filtered with 2 mm sieve and taken for reading.

General Terms Your general terms must be any term which can be used for general classification of the submitted material such as Pattern

Keywords Arsenic(As), NDSI, PLS, Reflectance Spectroscopy.

1. INTRODUCTION Soil pollution is one of the major problems in recent days. Pollution is caused by various Problems. It is particularly created by fertilizers, Chemical components, polluted water and other chemical elements such as Arsenic, copper, lead which cause environment loss. Soil fertility also lost due to soil pollution. There are various effects of soil pollution on human health. It reduces the growing crop growth and agriculture land [1]. The level of toxicity factors content in soils can also be calculated, but their happening based on large sampling and scientific analysis systems that requires extra time, not efficient, and luxurious when utilized on a tremendous scale. Spectral signature gives useful option to usual ways for soil factors analysis as a result of its correct, convenience and rapidity. With the recent time, copper and arsenic content in the soils of cultivation field increased quickly and depressed the quality of crops. Human life is having threat of excessive type of various pollution in soils and meals. The effective method used for monitoring of heavy metals in soils and evaluation of its impact on agriculture field. It is essential to develop new method which is fast convenient and useful [2] [3].

Figure 1: Study Area of Aurangabad District.

2.2 Spectral measurement of soil data and its Pre- processing The Fieldspec - 4 Spectroradiometer is used for collection of spectral signature of soil samples. The device captures spectral data having range from (350-1000nm㸧and 㸦10002500nm㸧and interpolated the information to construct 1-nmspaced data. Spectral signature is obtained with the wavelength range from 350 nm to 2500 nm [4]. The measured values are processed, and sooner or later the Spectroradiometer provides a spectrum of 2151 bands having a linear spectra interval of 1 nm. 7KHOLJKWVRXUFH¶VKHLJKWLVFm, height of Gun is 5 cm and exact distance between two devices light source and the gun is 50 cm. Samples are collected on 7/04/2016 and reflectance measured

2. MATERIALS AND METHODOLOGY 2.1 Study Area 396

Remote Sensing and GIS, Smart City and Smart Villages

3.2.Content of Arsenic in soil Using Normalized Difference Spectral Index 1'6,[\  \í[ [\ Where x and y are the high spectral signature values of soil VDPSOHVZLWKVSHFWUDOZDYHOHQJWKRIíQP NDSI (R812, R782) = (R812-R782) / (R812+R782) [6]

Figure 2: spectral signature of soil samples

2.3. Methodology

Figure 4: Arsenic Content in Each Sample. The arsenic content changes with different soil samples. With the help of normalized difference spectral index this content acquired. The range is between 0.28 to 1.02. Therefore soil pollution is correlated with toxicity of element which effects on soil quality and reduces crop growth. This is major concerned to crop damage. So the content level selects in what amount it matters on soil quality. The content level is stated with various stages that are varied from 0 to 0.5 are in safe approach then there is no risk to soil or crop. In between 0.5-1.0 range is particularly not as much of damaging and above it further injurious to soil quality. The below table shows the toxicity levels

4. CONCLUSION

Figure 3: Conceptual Model

3. EXPERIMENTAL ANALYSIS 3.1. Content of Arsenic Prediction in soil with PLSR The calculated accuracy strongly depends on the input data pretreatment method which is used to process the spectral signature. The calibration is carried out with the partial least squares (PLS) regression. The PLS is a valuable system for defining predictive models when there are various variables. It can use with multicollinearity and number of samples lower than predictor variables. All measurement is performed using Unscrambler X 10.4(64 bit) [5]. Table 1: Comparison of Processing Techniques

Processing Technique First Derivative Savitzky-Golay Median Filter Gaussian Filter

RMSE Cal 0.73 0.59 0.48 0.49

R2 Cal 0.95 0.92 0.95 0.95

This research contributes As content in soil samples with the help of spectral signature. Partial least square regression technique used for the identification of Arsenic content. The different processing method with the spectral data suggests good impact on the content prediction. The first Derivative is used for better noise removal on spectral data and to get the results. The high values with index having 782 and 812 nm is calculated from the spectral signature used as As contents prediction in soils.

5. ACKNOWLEDGMENT This work is supported by Department of Science and Technology under the Funds for Infrastructure under Science and Technology (DST-FIST) with sanction no. SR/FST/ETI340/2013 to Department of Computer Science and Information Technology, Dr. Babasaheb Ambedkar Marathwada University, Aurangabad, Maharashtra, India, the authors would like to thank Department and University Authorities for providing the infrastructure and necessary support for carrying out the research.

6. REFERENCES [1]

Guanghui Zheng and Shenglu Zhou,Guanghui Zhen, ³3RVVLELOLWLHV RI 5HIOHFWDQFH to Predict Soil $UVHQLF´,&,6( SS- 4992, Sept, 2009.

397

Cognitive Knowledge Engineering

[2]

Tiezhu SHI, Huizeng Liu, Junjie Wang, Yiyun Chen, 7HQJ)HL DQG *XRIHQJ :X ³0RQLWRULQJ $UVHQLF Contamination in Agriculture Soils with Reflectance 6SHFWURVFRS\RI5LFH3ODQWV´(QYLURQPHQWDO6FLHQFH& Technology),pp 6264-6472,May 14,2014.

[3]

REN Hong-Yan, ZHUANG Da-Fang, A. N. SINGH, PAN Jian-Jun, QIU Dong-6KHQJ ³(VWLPDWLRQ RI $V DQG &X Contamination in Agricultural Soils Around a Mining $UHD E\ 5HIOHFWDQFH 6SHFWURVFRS\  $ &DVH 6WXG\´ Elsevier Limited and Science Press, July 23, 2009.

[4]

Chang CW, Laird DA, Mausbach MJ, Hurburgh CRJ, ³1HDU-infrared reflectance spectroscopy-principal components regression analysis of VRLO SURSHUWLHV´ 6RLO Sci Soc Am J, pp- 480-490, 2001. Tiezhu SHI, Huizeng Liu, Junjie Wang, Yiyun Chen, 7HQJ)HL DQG *XRIHQJ :X ³0RQLWRULQJ $UVHQLF Contamination in Agriculture Soils with Reflectance 6SHFWURVFRS\RI5LFH3ODQWV´(QYLURQPHQWDO6FLHQFH  Technology),pp 6264-6472,May 14,2014.

[5]

[6]

@ $QMXJDP 0 .DZLWD 0 'U55  'U65   ³'HVLJQ andImplementation of Voice Control System For :LUHOHVV+RPH$XWRPDWLRQ1HWZRUNV´,&&&(,6%1 978-1-4675-2248-9.

>@ 2OXIHPL 6XQGD\ $GHR\H   ³$ 6XUYH\ RI (PHUJLQJ %LRPHWULF 7HFKQRORJLHV´ ,QWHUQDWLRQDO -RXUQDO RI Computer Applications (0975 ± 8887), Volume 9±No.10. [3] A. N. Mishra, Mahesh Chandra, Astik Biswas, S. N.Sharan,  ´ 5REXVW )HDWXUHV IRU &RQQHFWHG +LQGL 'LJLWV5HFRJQLWLRQ´ ,QWHUQDWLRQDO -RXUQDO RI 6LJQDO Processing, Image Processing and Pattern Recognition Vol. 4, No. 2. >@ 1HKD 'DKL\D 'U &KDQGHU .DQW   ³%LRPHWULFV 6HFXULW\ &RQFHUQV´ 6HFRQG ,QWHUQDWLRQDO &RQIHUHQFH RQ Advanced Computing & Communication Technologies, IEEE, pp. 297-302. >@ +LPDQVKX6HNKDU $FKDU\D   ³%LRPHWULFV LQ 0RGHUQ (UD´ International Journal of Advanced Research in Engineering and Applied Sciences ISSN: 2278-6252 Vol. 2,No. 3. [6] Bansod, N.S.; Dadhade, S.B.; Kawathekar, S.S.; Kale, K.V., (2014), "Speaker Recognition Using Marathi (Varhadi) Language," Intelligent Computing Applications (ICICA), 2014 International Conference on , vol., no., pp.421,425, 67. >@ 6 )XUXL   ³&HSVWUDO DQDO\VLV WHFKQLTXH IRU DXWRPDWLFVSHDNHU YHULILFDWLRQ´,((( 7UDQV $FRXVW Speech,Signal and Processing, vol. 29, pp.254272.Biometrics ³)RXQGDWLRQ'RFXPHQWV´1DWLRQDO6FLHQFH &Council P-92. >@ 16 %DQVRG 6% 'DEKDGHHWDO  ³5HYLHZ 2I 'LIIHUHQW7HFKQLTXHV )RU 6SHDNHU 5HFRJQLWLRQ 6\VWHP´ Advancesin Computational Research, ISSN: 0975-3273 & EISSN:0975-9085, Volume 4, Issue 1, pp.-57-60. >@$10LVKUDHWDO ³5REXVW)HDWXUHVIRU&RQQHFWHG +LQGL 'LJLWV 5HFRJQLWLRQ´ ,QWHUQDWLRQDO -RXUQDORI 6LJQDO Processing, Image Processing and Pattern Recognition, Vol. 4, No. 2. >@ 3DZDQ .XPDU HWDO  ³6SRNHQ /DQJXDJH IdentiILFDWLRQ 8VLQJ +\EULG )HDWXUH ([WUDFWLRQ 0HWKRGV´ Journal Of Telecommunications, Volume 1, Issue 2. >@ 3DZDQ .XPDU 0DKHVK &KDQGUD   ³6SHDNHU ,GHQWLILFDWLRQ 8VLQJ *DXVVLDQ 0L[WXUH 0RGHOV´ 0,7 International Journal of Electronics and Communication Engineering Vol. 1, No. 1, pp. (27-30), ISSN 22307672©MIT Publications. [12]Zaidi Razak, Noor Jamilah Ibrahim, emranmohdtamil,mohd