7. First applications and results. 7 pp o d. • 8. Future fusion schemes. Despini F.,
Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA ...
HSI 2012 Third Annual Hyperspectral Imaging Conference 15 – 16 May 2012, Rome, Italy
“Pan--Sharpening” “Pan p g procedures p for the hyperspectral sensor PRISMA Francesca F D Despini Despini, i i, Sergio S i T Teggii
DIMEC Department of Mechanical and Civil DIMEC, Engineering University of Modena and Reggio Emilia
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
OUTLINE • • • • • • • •
1. ASI-AGI Project 2. Data Fusion and Pan Sharpening 3. Process scheme 4 Image simulation 4. sim lation 5. Dataset 6. Data fusion methods 7. First applications 7 pp o and d results 8. Future fusion schemes Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
1 ASI-AGI Project 1. Analisi Sistemi Iperspettrali per l Applicazioni le li i i Geofisiche fi i h Integrate PRISMA (PRecursore Iperspettrale della Missione Applicativa) New platform of the Italian Space Agency project “ASI- AGI”
Main goal: Monitoring and understanding the Earth’s surface Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
1. ASI-AGI ASI AGI Project P j Main goals: • Realize a small national mission for the monitoring of natural resources and characteristics of the atmosphere; • Provide data with an high spectral resolution necessary to the scientific community to develop new applications for the observation of the territory and support the environmental risk management Principal p applications: pp • Environmental monitoring; • Inventory and monitoring of forests; • Geological mapping; • Land cover maps, inventory of crops; • Definition of the productivity of aquatic ecosystems of coastal and inland waters; • Monitoring of the carbon cycle; • Mapping of urban areas; • Characterization of the atmosphere; • Risk Ri k assesment Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
1 PRISMA 1. Hyperspectral sensor (HS) 0.4 – 2.5 μm (spectral resolution: 10 nm) Spatial resolution: 30m
Panchromatic sensor (PAN) 0.4 – 0.75 μm Spatial resolution: 5m
• The application of Pan Sharpening techniques to PRISMA allows ll considerable id bl advantages. d • The aim is to obtain new information with the spatial detail of the PAN and the spectral detail of HS bands Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
2. Data D Fusion F i and d Pan P Sharpening Sh i Data fusion
Pan Sharpening
Data fusion is “a formal framework in which are expressed means and tools for the alliance of data originating g g from different sources. It aims at obtaining information of a greater quality, although the exact definition of ‘greater quality’ will depend on the application application”(Wald, (Wald, 1999) It is the synthesis of hyperspectral images to the higher spatial resolution of the panchromatic image. The synthesis of fused HS images should be as close as possible to those that would have been observed if the corresponding sensors had the spatial resolution of the panchromatic sensor.
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
2. Data D Fusion F i and d Pan P Sharpening Sh i Advantages • An increasing number of applications such as feature detection de ec o o or land a d co cover e cclassification ass ca o require equ e high g spatial spa a a and d high spectral resolution at the same time for improved classification results, strengthened reliability, and/or a better visual i l interpretation. i t t ti • An high-quality g q y synthesis y of spectral p information is veryy important for most remote sensing applications based on spectral signatures, such as lithology and soil and vegetation analysis. analysis
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
3. Process scheme Input: • PRISMA simulated i l d HS image i ( (30 m) from the MIVIS sensor; • PRISMA simulated PAN image (5 m)) from f th MIVIS sensor; the • PRISMA simulated HS reference image (5 m)
Application of several fusion schemes: • Standardized PCA fusion scheme; Gram-Schmidt Schmidt fusion scheme; • Gram • Discrete Wavelet transform; • Hybrids methods.
Validation: • Correlation coefficient between the bands of the reference image and the same bands of the synthetic image; • Quantitative parameters: RMSE e ERGAS
Determination of the best methods for the fusion of groups of bands.
Output: • Synthetic HS image with an i improved d spatial ti l resolution l ti off 5 m;
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
4. Image I simulation i l i MIVIS Multispectral Infrared and Visible Imaging Spectrometer Optical Port
S Spectral t l range
VIS
0.431 – 0.833 μm
20
0.02 μm
NIR
1.150 – 1.550 μm
8
0.05 μm
SWIR
1.985 – 2.479 μm
64
0.008 μm
8.210 – 12.7 μm
10
0.40 μm
TIR
Number of Spectral channels resolution
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
4. Image I simulation i l i MIVIS Multispectral Infrared and Visible Imaging Spectrometer Advantages: Ad t - Better spatial resolution than PRISMA (validation) - Availability of images; - Total Spectral range similar to PRISMA. Disadvantages: - The spectral range is not continuous; - Different spectral resolution.
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
5. Dataset D Image of the city of Parma (Italy) Immagine utilizzata Acquired MIVIS on 11/06/1999 at 10:45:45 Spatial resolution: 1.64 1 64 m From this F hi HS image i ( bands) (92 b d ) we generated: - The PAN image with a spatial resolution of 5 m; g with a spatial p resolution of - The HS image 30 m; - The HS image with a spatial resolution of 5 m ((reference f iimage). ) Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
5. Dataset D HS image g (92 bands) 30 m
1
+
+
PAN image 5m
=
= Synthetic HS image (92 bands) 5m
1
2
2
3
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
3
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
6 Data 6. D fusion f i methods h d • •
Projection–Substitution All the HS bands are simultaneously synthesized. Main assumption: the structures contained in this systhesis are equivalent to those of PAN modality. y
•
Relative Spectral Contribution Main assumption: the low-resolution low resolution PAN can be written as a linear combination of original HS bandss. ARSIS Concept Main assumption: that the missing spatial information in the HS bands can be retrieved from the high g frequencies q components p of PAN
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
6 Data 6. D fusion f i methods h d Standard ized PCA ffusion scheme •
HS bands are transformed to their Principal Components (PC);
Hp: PC1 contains most of the data variance between all the bands, bands which which, in general, related to spatial information • • •
Linear Scaling of the PAN on the first Principal Component PC1; Replacement of PC1 with the scaled PAN.; Backward transformation to the new HS image.
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
6 Data 6. D fusion f i methods h d Gram – Schmidt ffusion scheme 1. Simulation of a PAN band at the lower spatial resolution using the HS bands 2. Gram-Schmidt transformation of simulated PAN and HS bands 3. Substitution of the first Gram-Schmidt band with high spatial resolution PAN 4. Backward Gram-Schmidt transformation to the new HS images
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
7. First Fi applications li i and d results l Ref. 5m
Standardized PCA ffusion scheme Synt. y Image • Excellent maintenance of spatial information ; • The correlation coefficient between the bands of the reference image and the same bands of the synthetic image varies from 0.59 to 0.99 Correlation Co C oefficient
1
C Correlation l i with i h the h reference f image i Correlation
0.9 0.8 0.7 0.6 0.5 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 Bands
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
7. First Fi applications li i and d results l Standardized PCA ffusion scheme
Reference 5m
Synthetic image
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
7. First applications 7 pp and results Gram-Schmidt fusion scheme Synt. Image g
Ref. 5m
Synt. g Image
Synt. Image g
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
1
2
3
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
7. First Fi applications li i and d results l Numeric global error parameter to describe the quality of the fusion method i: current pixel NP: number of pixels Bk: k-band of the hyperspectral reference image Bk*: k-band of the synthetic image h: spatial resolution of the panchromatic h i image i l: spatial resolution of the hyperspectral image Mk: mean radiance di for f each hb band d N: number of bands
Threshold value for ERGAS parameter = 3 Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
7. First Fi applications li i and d results l Standardized PCA ffusion scheme ERGAS
8.99
Spectral response 10000 9000
Reflectance
8000
Reference image
7000
Synthetic image
6000 5000 4000 3000 2000 1000 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91
Bands
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
7. First Fi applications li i and d results l Gram-Schmidt ffusion scheme ERGAS1
8.04
ERGAS2
8.11
ERGAS3
7 80 7.80
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
C Continuation i i off the h work k • The method test up to now did furnish satisfactory results • This is not surprising (literature): nevertheless this firs step was necessary to define the simulation methodologies and validation metrics (other metrics not ot sshown ow here e e aahss bee been considered) co s de ed) • Now we will start the study of different methods:
Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
F Future ffusion i schemes h Wavelet fusion schemes • Wavelet-based fusion schemes are extensions of the high-pass fil method, filter h d which hi h makes k use off the h idea id that h spatial i l detail d il is i contained in high frequencies. • In I the th wavelet-based l t b d fusion f i schemes, h detail d t il information i f ti is i extracted from the PAN image using wavelet transforms and injected into the HS image. • Distortion of the spectral information is minimized; however there mayy be other negative g effects. Some of these effects result from the type of wavelet transform that is used while others result from the method of injecting detail information i into the h MS S image. i Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 – Third Annual Hyperspectral Imaging Conference, 15 – 16 May 2012, Rome, Italy
F Future f i fusion schemes h Hybrid Methods • 1. Convert HS bands to PC;; • 2. Generate a new PAN image histogram-matched to PC1 component; • 3. Apply DWT (Discrete Wavelet Transform ) to new PAN and PC1 images; • 4. Apply selected model (substitution, addition, etc..) to obtain set of approximation and detail images; • 5. Perform inverse DWT to obtain fused PC1 component; • 6. Convert new PC to HS bands. Despini F., Teggi S.– “Pan-Sharpening” procedures for the hyperspectral sensor PRISMA
HSI 2012 Third Annual Hyperspectral Imaging Conference 15 – 16 May 2012, Rome, Italy
Thanks Th k for f your kind ki d attention!
[email protected] [email protected]
DIMEC Department of Mechanical and Civil DIMEC, Engineering University of Modena and Reggio Emilia