Waste Management 56 (2016) 46–52
Contents lists available at ScienceDirect
Waste Management journal homepage: www.elsevier.com/locate/wasman
An investigation of the usability of sound recognition for source separation of packaging wastes in reverse vending machines M. Kemal Korucu a,⇑, Özgür Kaplan b, Osman Büyük c, M. Kemal Güllü c a
Kocaeli University, Department of Environmental Engineering, 41380 Kocaeli, Turkey Kocaeli University, Department of Mechanical Engineering, 41380 Kocaeli, Turkey c Kocaeli University, Department of Electronics and Communications Engineering, 41380 Kocaeli, Turkey b
a r t i c l e
i n f o
Article history: Received 2 November 2015 Revised 26 May 2016 Accepted 23 June 2016 Available online 1 July 2016 Keywords: Hidden Markov model Municipal solid wastes Reverse vending machines Sound of packaging wastes Support vector machines Voice recognition
a b s t r a c t In this study, we investigate the usability of sound recognition for source separation of packaging wastes in reverse vending machines (RVMs). For this purpose, an experimental setup equipped with a sound recording mechanism was prepared. Packaging waste sounds generated by three physical impacts such as free falling, pneumatic hitting and hydraulic crushing were separately recorded using two different microphones. To classify the waste types and sizes based on sound features of the wastes, a support vector machine (SVM) and a hidden Markov model (HMM) based sound classification systems were developed. In the basic experimental setup in which only free falling impact type was considered, SVM and HMM systems provided 100% classification accuracy for both microphones. In the expanded experimental setup which includes all three impact types, material type classification accuracies were 96.5% for dynamic microphone and 97.7% for condenser microphone. When both the material type and the size of the wastes were classified, the accuracy was 88.6% for the microphones. The modeling studies indicated that hydraulic crushing impact type recordings were very noisy for an effective sound recognition application. In the detailed analysis of the recognition errors, it was observed that most of the errors occurred in the hitting impact type. According to the experimental results, it can be said that the proposed novel approach for the separation of packaging wastes could provide a high classification performance for RVMs. Ó 2016 Elsevier Ltd. All rights reserved.
1. Introduction Municipal solid waste (MSW) management is a multidisciplinary activity that includes generation, source separation, collection, transportation, treatment, and disposal (Rada et al., 2013; Das and Bhattacharyya, 2015). One of the most important objectives of MSW management applications is the separation of recyclable packaging wastes at source with a high efficiency. For example, Turkey government recently set a 60% target value for the separate collection of the packaging wastes (LCPW, 2011). This target has not been realized in Turkey, yet. On the other hand, some European countries have accomplished 80% efficiency ratio for the separate collection of all waste fractions (Rada et al., 2013). Source separation of MSWs refers to the separate collection of different waste fractions such as metal, glass, plastic, and cardboard, in the place where they are generated (Rousta et al., 2015). Nowadays, the source separation of recyclable wastes is ⇑ Corresponding author at: Kocaeli University, Umuttepe Campus, Engineering Faculty, Environmental Engineering Department, 41380 Kocaeli, Turkey. E-mail address:
[email protected] (M.K. Korucu). http://dx.doi.org/10.1016/j.wasman.2016.06.030 0956-053X/Ó 2016 Elsevier Ltd. All rights reserved.
accomplished via deposit applications (e.g. the discount tickets for empty bottles, etc.), colored bags and separate collection containers (Dahlén et al., 2007; Larsen et al., 2010; Gallardo et al., 2012; Lavee and Nardiya, 2013; Groot et al., 2014; Rigamonti et al., 2014; Zhang and Wen, 2014). However, there is always a risk of consumer misuse in these conventional applications: The consumer can throw the waste to a wrong bag and/or an incorrect part of the container. Reverse vending machine (RVM) is a device used for the separate collection of packaging wastes automatically. These devices can successfully disable human factors in the source separation. They prevent aforementioned human misuse substantially, reward the users directly, and punish misuse when necessary. Working principles of RVMs can be listed as; 1 – proximity sensors (Van Den Broek et al., 1997, 1998), 2 – image processing (Ramli et al., 2007), 3 – barcode reading (Wyld, 2010), and 4 – radio frequency identification (RFID) (Thomas, 2008; Glouche and Couderc, 2013). The first three techniques are inefficient in waste management applications due to the vast variety of packaging wastes in terms of fullness, deformation conditions, shape, structure and mass. Tagging all the packaging materials with an RFID tag during the
M.K. Korucu et al. / Waste Management 56 (2016) 46–52
manufacturing phase might seem to be an efficient solution. However, this tagging process significantly increases the costs (Binder et al., 2008; Thomas, 2009: 0.05–0.10 $/tag) and the tags are harmful to the environment. For example, if the RFID technology is started to be used worldwide today, approximately two trillion products per year should be tagged (Thomas, 2008). These tags could contain toxic or valuable materials (Abdoli, 2009). The possibility of releasing the toxic tags to the environment, the damage they cause in the separation automats (Wager et al., 2005) and the operational problems (Aliaga et al., 2011) restrict the wide usage of the RFID technique. Consequently, it is obvious that a new efficient and cheap source separation technique for RVMs should be developed to overcome the weaknesses of the current technologies. In this study, we investigate the use of sound recognition for source separation of packaging wastes in RVMs and present the initial experimental results. In the experiments, sounds were generated from packaging wastes via free falling, pneumatic hitting and hydraulic crushing impact types. The sounds were recorded with two different microphones. A support vector machine (SVM) and a hidden Markov model (HMM) based classification systems were trained using the sound recordings. The systems were tested with packaging wastes in different material type and size. In the experiments, it was observed that the classification systems provide high classification accuracies (between 82% and 100%) for the separation of packaging wastes.
2. Theory, material and methods 2.1. Theory It is possible to identify the type (glass, metal, plastic, etc.) and/or shape of a solid material using the sounds reflected from it or generated by it. Zhao et al. (2003), Ohtani and Baba (2006a,b), Gonzalez et al. (2011, 2012) and Zhang et al. (2013). Characteristic of a sound reflected or generated from a solid material is affected by factors such as the shape and type of the material, and the type of the physical impact (Avanzini and Rocchesso, 2001). It has been proven that sound generator objects could be identified from their shapes (Kunkler-Peck and Turvey, 2000), sizes (Giordano and McAdams, 2006) and material structures (Giordano, 2003; McAdams et al., 2004, 2010) by humans. According to Tucker and Brown (2002), the most distinguishing feature of a sound generator is its material structure for a human listener. In Rocchesso and Fontana (2003), it is stated that the research on identifying an object from its sound feature started in the 1970s and many algorithms have been developed in this research area. For example, Wildes and Richards (1988) developed two theoretical statements namely ‘‘bandwidth” and ‘‘decay” to identify the material structure of non-elastic materials. Lutfi and Oh (1997) and Klatzky et al. (2000) performed experiments using the collision sounds to assess the material identification capacity of human listeners. Avanzini and Rocchesso (2001) used a hammer to generate a sound. Besides the above researches; sound/speech processing technologies has evolved significantly. Today, it is an accomplished technological achievement to classify an object using the sound signals generated and/or reflected from it (Madisetti, 2009; Lyons, 2010; Ingle and Proakis, 2011). Various algorithms and computer tools have been developed for this purpose (Young et al., 2006; Sphinx, 2009). It could be found in the studies of Haff and Pearson (2007), Ocak (2009), Chang and Lai (2010), Madain et al. (2010), Subha et al. (2010), Tran and Li (2011), El-Alfi et al. (2013), Guyot et al. (2013), Yuan and Ramli (2013) and Theodorou et al. (2015) that various feature types and classification methods can be used to classify the sounds with different characteristics.
47
Just like the other solid materials, packaging wastes generate different sounds under a physical impact. However, to the best knowledge of the authors, there is no previous study in the literature that uses sound recognition to identify the packaging wastes. In this study, we focus on using sound recognition approach for separate collection of the packaging materials in RVMs. The main steps of the methodological approach used in this study can be found in Fig. 1. 2.2. Experimental setup and sound recording procedure An open-top cubic structured steel chamber with 50 cm 50 cm 50 cm dimensions was designed for the sound recording experiments. The chamber was isolated with rock wool panels. The thickness of the isolation was 10 cm. All the waste sound recordings were taken inside this chamber. The recordings were obtained with the help of the free falling (FF), pneumatic hitting (H) and hydraulic crushing (C) physical impact types. Pneumatic and hydraulic cylinders were used for the H and C type impacts, respectively. The maximum allowable pressure of the hydraulic system was 40 bars. On the other hand, the pneumatic system was operated under a constant pressure of four bars. A pressure regulator was used to maintain the constant pressure. The experimental setup and some of the packaging wastes used in this study can be seen in Fig. 2. Packaging wastes used in this study were collected from student canteens in Kocaeli University and shopping malls in Kocaeli city center. All the collected wastes were empty and they had been consumed and dumped to trash bins by consumers. The wastes were categorized into four groups according to the material type; metal (M), plastic (P), glass (G), and cardboard (CB). As mentioned earlier, RVMs are high-tech machines which classify the waste given by the consumer and reward the consumer according to the waste type. Therefore, the size of the wastes should also be considered for the rewarding mechanism in the real world waste management applications. For this reason, the wastes were collected in three different sizes for every waste category above, namely big (B), medium (MS) and small (S). The sounds were obtained for the three physical impact types; free falling (FF), pneumatic hitting (H) and hydraulic crushing (C). Sound recordings were taken with two different microphones. One of the microphones is a dynamic (D) and the other is a condenser (CN) microphone. Technical specifications of the dynamic microphone are 40 Hz–15 kHz frequency response, 600 ohm impedance, and 54 dB re one V/Pa sensitivity for MXL LSM-5GR. Technical specifications of the condenser microphone are 20 Hz–20 kHz frequency response, 150 ohm impedance, and 30 dB re one V/Pa sensitivity for MXL CR89. The sounds were recorded on a computer via a sound card using Cubase5 software. Also a video recording was taken for possible future research. The sound recordings were taken for every waste type-waste size-impact type combination, individually. A coding scheme was developed in order to present the experimental results more clearly. In the coding scheme, the following order was used: ‘‘waste type _waste size_impact type_experiment number_microphone type”. The total number of the waste types used in this study was 11 due to the absence of medium size metal waste. These waste types and their specifications can be found in Table 1. Some of the experimental conditions are also presented in the table for a better understanding of the coding scheme. In this study, different brands and various deformation conditions of waste types were taken into account in order to make a more realistic research. In real world applications, the compositions of the waste types might differ from one brand to another and they might be deformed when dumped to the trash bins. As a result, five different brands and three
48
M.K. Korucu et al. / Waste Management 56 (2016) 46–52
Analyzing the recordings and developing the sound recognition modeling systems
Collecting of used and empty glass, plastic, metal and cardboard packaging wastes
Design of Experimental Set-up
Collecting of Package Wastes
Basic Set-up Condition for SVM and HMM Comparison: - Free falling impact type
Sound Recording Studies
Sound Recognition Studies
Results and Discussion
Recording of the generated sounds from the wastes in the experimental set-up
Expanded Set-up Condition for HMM Application: - Free falling, pneumatic hitting and hydraulic crushing impact type Fig. 1. Methodological approach used in this study.
3
3
2 4 1
2
1: Pneumatic hitting cylinder 2: Camera 3: Microphones (dynamic and condenser) 4: Hydraulic crushing cylinder Fig. 2. Experimental setup and some packaging wastes used in this study.
49
M.K. Korucu et al. / Waste Management 56 (2016) 46–52 Table 1 Experimental coding scheme used in this study. Samples
Coding system
Acronym
‘‘ALU-41” type materials between 200 and 500 mL (packaging wastes of fruit juices, beer, beverages, etc.)
Big metal Small metal
M_B M_S
‘‘GL-70”, ‘‘GL-71” and ‘‘GL-72” type materials between 200 and 1000 mL (packaging wastes of mineral water, beer, wine, etc.)
Big glass Medium glass Small glass
G_B G_MS G_S
‘‘PET-1” type materials between 200 and 2000 mL (packaging wastes of drinking water, beverages, etc.)
Big plastic Medium plastic Small plastic
P_B P_MS P_S
‘‘C/PAP-84” type materials between 200 and 1000 mL (packaging wastes of milk, fruit juices, etc.)
Big cardboard Medium cardboard Small cardboard
CB_B CB_MS CB_S
Examples for coding system Metal Waste _ Big Size _ Free Falling _ Experiment No:1 _ Dynamic Microphone
M_B_FF_1_D
Cardboard _ Medium Size _ Hydraulic Crushing _ Experiment No:4 _ Dynamic Microphone
CB_MS_C_4_D
Glass Waste _ Small Size _ Pneumatic Hitting _ Experiment No:7 _ Condenser Microphone
G_S_H_7_CN
different deformation conditions (i.e., unstrained, crooked, highly crooked) were used in the experiments. Under these circumstances the total number of packaging wastes used in this study was 165. In the sound recording phase of the study, every waste sample was dropped so as to free fall from 30 cm height above impact surface for the free falling impact type (FF). The sound generated from FF impact type was recorded with the microphones. Following the FF impact type, the landed waste was hit by the pneumatic cylinder (H) and the generated sound was recorded. After the H impact type, the waste was crushed with the hydraulic cylinder (C) to record the sound. All the sound samples were recorded in 44,100 Hz sampling rate with 16 bits quantization level. The recordings were taken in uncompressed pulse code modulation (PCM) format. In the sound recognition experiments of this study, it was observed that hydraulic crushing (C) impact type recordings were very noisy. These recordings need additional noise removal preprocessing stage to obtain comparable results to the other two impact types. Additionally, the setup for C type is more expensive when compared to the other two impact types. To make a cheaper automatic classification system for RVMs, the free falling and pneumatic hitting impact types should be considered first. Therefore, the results of hydraulic crushing impact type were not presented in the following sections. As a result, the sound recordings in this study were evaluated for two different impact types (free falling and pneumatic hitting) using two different microphones. Considering the number of waste, the total number of recordings used in this study was 660 (165 waste types two impact types two microphones). 2.3. Sound recognition procedure Sound recognition studies mainly consist of three phases: feature extraction, model training and model testing (Gaikwad et al., 2010). There are numerous techniques available in the literature for each phase. For example, mel-frequency cepstral coefficient (MFCC), linear predictive coding coefficients (LPCC), and wavelet transform (WT) (Hibare and Vibhute, 2014) are some of the feature extraction methods. Modeling approaches can be classified into two categories: generative models and discriminative models (Kinnunen and Li, 2010; Gaikwad et al., 2010). Support vector machine (SVM) is a discriminative model specifically used for bin-
ary pattern classification problems, but it can also be extended to a multi-class problem (Yang et al., 2013). HMM is a generative model which has found various applications in pattern recognition problems such as speech recognition, speaker recognition, hand-writing analysis and gesture recognition (Gales and Young, 2008; Shahin, 2008; He et al., 2008; Elmezain et al., 2009). SVM and HMM based classification approaches with MFCC features have been widely used for sound recognition (Ananthi and Dhanalakshmi, 2015). In this study, two sound classification systems were developed based on SVM and HMM models. In the first phase of the study, SVM and HMM based classifications were compared on just the free falling data (i.e., basic condition of the setup) to evaluate the appropriateness of the sound recognition to the packaging waste separation problem in the cheapest way. In this phase, the examinations were made in terms of material type classification. In the second phase of the study, material sizes were also considered in addition to the material type. The classification were performed by the HMM classifier because any silence/utterance alignment that would simplify the classification of the system was not performed for the SVM system. In this phase, the sound recordings of three impact type were used together (i.e., expanded condition of the setup). In the classification studies, 13 MFCCs and their first order derivatives were extracted from the signals for each 25 ms time window with 10 ms overlapping. Therefore, 26 features were obtained for each window. Cepstral mean subtraction was applied to obtain final feature vectors. Training and testing of the SVM classifier were performed in MATLAB software using LibSVM library (Chang and Lin, 2011). Multi-class classification was carried out using one-versus-one strategy to determine the output prediction. Six different classifiers were trained to separate glass-cardboard, glass-metal, glass-plastic, cardboard-metal, cardboard-plastic, and metal-plastic classes. Parameters were determined by 10fold cross validation, and 90%, 80%, 70%, 60%, 50% and 10% training levels were utilized. For HMM classifier, separate HMMs were trained for each waste and impact types. Additionally, a HMM training was performed in order to align the silent segments in the beginning and ending of the recordings. Each model had left-to-right HMM topology with three emitting states and four mixtures. For the recognition experiment, a simple finite state grammar which consists of 22 classification classes (11 different waste types and two different impact type) was constructed. The
50
M.K. Korucu et al. / Waste Management 56 (2016) 46–52
Table 2 Classification accuracies obtained in the study. Training level of model (%)
Classification accuracy of model for dynamic microphone (%)
Classification accuracy of model for condenser microphone (%)
10
85
82
50
97.3
90.4
60
95
91.7
70
95.7
95.7
80
96.7
93.3
90
100
100
90
100
100
Material type identification of HMM classifier in the expanded conditions of the experimental setup
96.5
97.7
Material type and size identification of HMM classifier in the expanded conditions of the experimental setup
88.6
88.6
Material type identification of SVM classifier in the basic conditions of the experimental setup
Material type identification of HMM classifier in the basic conditions of the experimental setup
best matching grammar item to the test recording was searched using the Viterbi algorithm for the classification decisions. The HMM based system was realized with hidden Markov model toolkit (HTK) (Young et al., 2006). Ninety percent of the recordings were used in the model training, and 10% of them were used in the testing. In training phase, most of the recordings were used for the HMM training in order to obtain robust HMM parameters.
3. Results and discussion Classification accuracies (CAs) obtained in the study are presented in Table 2. As it can be seen from Table 2, SVM classifier provided 100% accuracy for free falling data obtained by both dynamic and condenser microphones in 90% training level. Additionally, for different training ratios, SVM classifier provided different CA values for dynamic and condenser microphones. According to these results, it can be stated that SVM classifier provides higher recognition ratios with dynamic microphone for the free falling dataset. Hence, the dynamic microphone seems more suitable compared to the condenser one if we use SVM classifier. The HMM classifier was evaluated for 90% training ratio, and obtained CA was 100% for both microphones for free falling test, similar to the SVM. In the expanded condition of the setup, on the other hand, CA values of the HMM classifier were 96.5% and 97.7% for dynamic and condenser microphone, respectively. Additionally, CA values of the HMM classifier for the simultaneous classification of material type and size were 88.6% for both microphones. Table 3 Misclassified test results for the HMM classifier. Error No.
Test input
Decision
1
CB_B_H_D
CB_MS_H_D
Failure Size
2
CB_S_H_D
CB_MS_H_D
Size
3
CB_MS_H_D
P_B_H_D
Type and size
4
CB_MS_H_D
P_B_H_D
Type and size
5
P_B_H_D
P_MS_H_D
Size
6
CB_B_FF_CN
CB_MS_FF_CN
Size
7
CB_B_H_CN
CB_MS_H_CN
Size
8
CB_MS_H_CN
P_B_H_CN
Type and size
9
CB_MS_H_CN
CB_B_H_CN
Size
10
P_B_H_CN
P_S_H_CN
Size
Misclassifications of HMM model were detected on the 10 of 88 test records which are presented in Table 3 (please note that there is no additional sound recording and/or modeling attempt to improve the results). As it can be seen from Table 3, the HMM classifier provided the same number of misclassification for both microphones. It is known that the condenser microphones have a wider range of frequency response, but they are more expensive then dynamic ones. Since the aim of this work is to develop an efficient and cheap separate collection technique, dynamic microphone can be suggested in further studies, keeping in mind the aforementioned CA values. According to Table 3, 90% of the misclassifications were caused by the hitting impact type. In this sense, free falling impact type which is the most simple and cheap impact type of this study can be easily suggested for future study. On the other hand, further studies should certainly be conducted with hydraulic crushing impact type. Since crushing a material increases the amount of waste which can be stored in the collection bin dramatically, a double advantage may be provided via the possible improvements of using this type impact in RVMs. As it is shown in Table 3, only two of 44 test recordings of the dynamic microphone and one of 44 test recordings of the condenser microphone were misclassified in terms of material type identification of the wastes. Cardboard wastes were identified as plastic in all these misclassifications. Furthermore, the results of SVM classifier indicated that misclassification ratio for the plastic wastes proportionally increases if training ratio is decreased. These misclassifications were mostly associated with glass wastes in the way that identification of a glass as a plastic or vice versa. All in all, it is aimed to investigate the misclassification of the plastic wastes in future studies. The lowest classification accuracy of the HMM classifier was 96.5% for the material type classification, and this accuracy is a promising result for RVMs. 88.6% accuracy on the simultaneous classification of material type and size is also remarkable for this first experiments. There is unfortunately no data is available in the literature for barcode reading and RFID based RVMs. Therefore, we have no chance to compare the classification results. The obtained CA values in this study indicate that even if it is desired to identify both the material type and size, this approach can provide high separation accuracies. Furthermore, it may be possible to classify sub-material types (e.g., FE-40, ALU-41, GL-70, GL-71, etc.) with the help of this sound based separation technique. One important limitation of this study was the relatively small number of waste samples used in the experiments. Furthermore, all the packaging materials used in this study are empty and only SVM and HMM classifiers are evaluated in this study. Further
M.K. Korucu et al. / Waste Management 56 (2016) 46–52
research should be included half empty and full packaging wastes in order to build a sophisticated waste separation method. Additionally, numerous classification methods are available in the literature that can be investigated. And possible combinations of sound and image classification approaches in RVMs can be considered in future studies. Acknowledgments This study is financed by the Scientific Research Projects Unit (KOU BAP) of Kocaeli University. References Abdoli, S., 2009. RFID application in municipal solid waste management system. Int. J. Environ. Res. 3 (3), 447–454. Aliaga, C., Ferreira, B., Hortal, M., Pancorbo, M.Á., López, J.M., Navas, F.J., 2011. Influence of RFID tags on recyclability of plastic packaging. Waste Manage. 31 (6), 1133–1138. Ananthi, S., Dhanalakshmi, P., 2015. SVM and HMM modeling techniques for speech recognition using LPCC and MFCC features. In: Proceedings of the 3rd International Conference on Frontiers of Intelligent Computing: Theory and Applications (FICTA), pp. 519–526. Avanzini, F., Rocchesso, D., 2001. Controlling material properties in physical models of sounding objects. In: Proceedings of the International Computer Music Conference, Cuba. Binder, C.R., Quirici, R., Domnitcheva, S., Staubli, B., 2008. Smart labels for waste and resource management. J. Ind. Ecol. 12 (2), 207–228. Chang, G.C., Lai, Y.F., 2010. Performance evaluation and enhancement of lung sound recognition system in two real noisy environments. Comput. Methods Prog. Biomed. 97 (2), 141–150. Chang, C.C., Lin, C.J., 2011. LIBSVM: a library for support vector machines. ACM Trans. Intel. Syst. Technol. 2 (3), 27. Dahlén, L., Vukicevic, S., Meijer, J.E., Lagerkvist, A., 2007. Comparison of different collection systems for sorted household waste in Sweden. Waste Manage. 27 (10), 1298–1305. Das, S., Bhattacharyya, B.K., 2015. Optimization of municipal solid waste collection and transportation routes. Waste Manage. 43, 9–18. El-Alfi, A.E., Elgamal, A.F., Ghoniem, R.M., 2013. A computer-based sound recognition system for the diagnosis of pulmonary disorders. Int. J. Comput. Appl. 66 (17), 22–30. Elmezain, M., Al-Hamadi, A., Appenrodt, J., Michaelis, B., 2009. A hidden Markov model-based isolated and meaningful hand gesture recognition. Int. J. Electr. Comput. Syst. Eng. 3 (3), 156–163. Gaikwad, S.K., Gawali, B.W., Yannawar, P., 2010. A review on speech recognition technique. Int. J. Comput. Appl. 10 (3), 16–24. Gallardo, A., Bovea, M.D., Colomer, F.J., Prades, M., 2012. Analysis of collection systems for sorted household waste in Spain. Waste Manage. 32 (9), 1623– 1633. Gales, M., Young, S., 2008. The application of hidden Markov models in speech recognition. Found. Trends Signal Process. 1 (3), 195–304. Giordano, B.L., 2003. Material categorization and hardness scaling in real and synthetic impact sounds. In: Rocchesso, D., Fontana, F., (Eds.), The Sounding Object, pp. 73–94. Giordano, B.L., McAdams, S., 2006. Material identification of real impact sounds: effects of size variation in steel, glass, wood, and plexiglass plates. J. Acoust. Soc. Am. 119 (2), 1171–1181. Glouche, Y., Couderc, P., 2013. A smart waste management with self-describing objects. In: The Second International Conference on Smart Systems, Devices and Technologies (SMART’13), Italy. Gonzalez, J.J., Jimenez, J.A., Guzman, J.A., 2011. Peniel method for the ultrasonic material identification in robots. In: 2011 IEEE International Symposium Signal Processing and Information Technology (ISSPIT), pp. 496–501. Gonzalez, J.J., Jimenez, J.A., Guzman, J.A., 2012. Design and Construction of An Ultrasonic Sensor for the Material Identification in Robotic Agents. INTECH Open Access Publisher. Groot, J., Bing, X., Bos-Brouwers, H., Bloemhof-Ruwaard, J., 2014. A comprehensive waste collection cost model applied to post-consumer plastic packaging waste. Resour. Conserv. Recycl. 85, 79–87. Guyot, P., Pinquier, J., André-Obrecht, R., 2013. Water sound recognition based on physical models. In: 2013 IEEE International Conference Acoustics, Speech and Signal Processing (ICASSP), pp. 793–797. Haff, R.P., Pearson, T.C., 2007. Separating in-shell pistachio nuts from kernels using impact vibration analysis. Sens. Instrum. Food Qual. Saf. 1, 188–192. He, Z., You, X., Tang, Y.Y., 2008. Writer identification of Chinese handwriting documents using hidden Markov tree model. Pattern Recogn. 41 (4), 1295– 1307. Hibare, R., Vibhute, A., 2014. Feature extraction techniques in speech processing: a survey. Int. J. Comput. Appl. 107 (5).
51
Ingle, V., Proakis, J., 2011. Digital Signal Processing Using Matlab. Cengage Learning, Boston. Kinnunen, T., Li, H., 2010. An overview of text-independent speaker recognition: from features to super vectors. Speech Commun. 52 (1), 12–40. Klatzky, R.L., Pai, D.K., Krotkov, E.P., 2000. Perception of material from contact sounds. Presence-Teleop. Virt. 9 (4), 399–410. Kunkler-Peck, A.J., Turvey, M.T., 2000. Hearing shape. J. Exp. Psychol. – Hum. Percept. Perform. 26 (1), 279–294. LCPW, 2011. Legislation for Control of the Packaging Waste. Turkish Ministry of Environment and Forestry, Ankara, Turkey. Larsen, A.W., Merrild, H., Møller, J., Christensen, T.H., 2010. Waste collection systems for recyclables: an environmental and economic assessment for the municipality of Aarhus (Denmark). Waste Manage. 30 (5), 744–754. Lavee, D., Nardiya, S., 2013. A cost evaluation method for transferring municipalities to solid waste source-separated system. Waste Manage. 33 (5), 1064–1072. Lutfi, R.A., Oh, E.L., 1997. Auditory discrimination of material changes in a struckclamped bar. J. Acoust. Soc. Am. 102 (6), 3647–3656. Lyons, R.G., 2010. Understanding Digital Signal Processing. Pearson Education. Madain, M., Al-Mosaiden, A., Al-khassaweneh, M., 2010. Fault Diagnosis in Vehicle Engines Using Sound Recognition Techniques. 2010 IEEE International Conference Electro/Information Technology (EIT), pp. 1–4. Madisetti, V., 2009. Digital Signal Processing Fundamentals. CRC Press. McAdams, S., Chaigne, A., Roussarie, V., 2004. The psychomechanics of simulated sound sources: material properties of impacted bars. Acoust. Soc. Am. 115 (3), 1306–1320. McAdams, S., Chaigne, A., Roussarie, V., Giordano, B.L., 2010. The psychomechanics of simulated sound sources: material properties of impacted thin plates. Acoust. Soc. Am. 128 (3), 1401–1413. Ocak, H., 2009. Automatic detection of epileptic seizures in EEG using discrete wavelet transform and approximate entropy. Exp. Syst. Appl. 36 (2), 2027– 2036. Ohtani, K., Baba, M., 2006a. A simple identification method for object shapes and materials using an ultrasonic sensor array. In: Instrumentation and Measurement Technology Conference, IMTC 2006, pp. 2138–2143. Ohtani, K., Baba, M., 2006b. An identification approach for object shapes and materials using an ultrasonic sensor array. In: International Joint Conference SICE-ICASE, pp. 1676–1681. Ramli, S., Mustafa, M.M., Hussain, A., Wahab, D.A., 2007. Automatic detection of ’ROIs’ for plastic bottle classification. In: 5th Student Conference Research and Development SCORED, pp. 1–5. Rada, E.C., Ragazzi, M., Fedrizzi, P., 2013. Web-GIS oriented systems viability for municipal solid waste selective collection optimization in developed and transient economies. Waste Manage. 33 (4), 785–792. Rigamonti, L., Grosso, M., Møller, J., Sanchez, V.M., Magnani, S., Christensen, T.H., 2014. Environmental evaluation of plastic waste management scenarios. Resour. Conserv. Recycl. 85, 42–53. Rocchesso, D., Fontana, F., 2003. The Sounding Object, Mondo Estremo. Rousta, K., Bolton, K., Lundin, M., Dahlén, L., 2015. Quantitative assessment of distance to collection point and improved sorting information on source separation of household waste. Waste Manage. 40, 22–30. Shahin, I., 2008. Speaker identification in the shouted environment using suprasegmental hidden Markov models. Signal Process. 88 (11), 2700–2708. Sphinx, 2009. Carnegie Mellon University Sphinx, Open Source Toolkit for Speech Recognition http://cmusphinx.sourceforge.net/. Subha, D.P., Joseph, P.K., Acharya, R., Lim, C.M., 2010. EEG signal analysis: a survey. J. Med. Syst. 4 (2), 195–212. Theodorou, T., Mporas, I., Fakotakis, N., 2015. Automatic sound recognition of urban environment events. In: Speech and Computer. Springer International Publishing, pp. 129–136. Thomas, V.M., 2008. Environmental implications of RFID. In: IEEE International Symposium Electronics and the Environment ISEE 2008, pp. 1–5. Thomas, V.M., 2009. A universal code for environmental management of products. Resour. Conserv. Recycl. 53 (7), 400–408. Tran, H.D., Li, H., 2011. Sound event recognition with probabilistic distance SVMs. Audio Speech Lang. Process. 19 (6), 1556–1568. Tucker, S., Brown, G.J., 2002. Investigating the perception of the size, shape and material of damped and free vibrating plates. In: University of Sheffield, Department of Computer Science, Technical Report CS-02-10. Van den Broek, W., Wienke, D., Melssen, W.J., Feldhoff, R., Huth-Fehre, T., Kantimm, T., Buydens, L.M.C., 1997. Application of a spectroscopic infrared focal plane array sensor for on-line identification of plastic waste. Appl. Spectrosc. 51 (6), 856–865. Van Den Broek, W., Wienke, D., Melssen, W.J., Buydens, L.M.C., 1998. Plastic material identification with spectroscopic near infrared imaging and artificial neural networks. Anal. Chim. Acta 361, 161–176. Wager, P.A., Eugster, M., Hilty, L.M., Som, C., 2005. Smart labels in municipal solid waste - a case for the precautionary principle? Environ. Impact Asses. Rev. 25 (5), 567–586. Wildes, R.P., Richards, W.A., 1988. Recovering material properties from sounds. In: Richards, W.A. (Ed.), Natural Computation. MIT Press, pp. 356–363. Wyld, D.C., 2010. Taking out the trash (and the recyclables): RFID and the handling of municipal solid waste. Int. J. Software Eng. Appl. 1 (1), 1–13.
52
M.K. Korucu et al. / Waste Management 56 (2016) 46–52
Yang, X., Yu, Q., He, L., Guo, T., 2013. The one-against-all partition based binary tree support vector machine algorithms for multi-class classification. Neurocomputing 113, 1–7. Young, S., Evermann, G., Gales, M., Hain, T., Kershaw, D., Liu, X., Moore, G., Odell, J., Ollason, D., Povey, D., Valtchev, V., Woodland, P., 2006. In: The HTK Book (for HTK Version 3.4). Cambridge University Engineering Department. Yuan, C.L.T., Ramli, D.A., 2013. Frog sound identification system for frog species recognition. In: Context-Aware Systems and Applications. Springer, Berlin Heidelberg, pp. 41–50.
Zhang, B., Chen, K., Liang, Y., 2013. Analysis on reliability of acoustic features for material recognition of damped impacted plates. In: 2013 International Conference Information Science and Technology (ICIST), pp. 1641–1644. Zhang, H., Wen, Z.G., 2014. The consumption and recycling collection system of PET bottles: a case study of Beijing, China. Waste Manage. 34 (6), 987–998. Zhao, B., Basir, O.A., Mittal, G.S., 2003. Detection of metal, glass and plastic pieces in bottled beverages using ultrasound. Food Res. Int. 36 (5), 513–521.
