a comparison of unmixing algorithms for hyperspectral imagery · in this poster, we present an...
TRANSCRIPT
In this poster, we present an experimental comparison of unmixing using the constrained positive matrix factorization (cPMF) developed by researchers at CenSSIS with standard unmixing algorithms such as SMACC and MaxD that retrieve endmembers from the image. Experimental results using AISA hyperspectral images collected over Vieques Island in Puerto Rico are presented. The retrieved signatures have similar spectral features but the slight differences in the spectra retrieved by cPMF, in most cases, resulted in abundance maps that better describe the expected spatial distribution of the endmembers. Field work, ancillary data, and expert knowledge of the area are used to evaluate the results.
A Comparison of Unmixing Algorithms for Hyperspectral ImageryA Comparison of Unmixing Algorithms for Hyperspectral ImageryAndrea Santos-García1, Dr. Miguel Vélez Reyes – Advisor 1 , MSc. Samuel Rosario1 ,Dr. Danilo Chinea2
E-mail: [email protected], [email protected], [email protected], [email protected] 1 Laboratory for Applied Remote Sensing and Image Processing, CenSSIS University of Puerto Rico-Mayaguez, PR.
2 Biology Department, University of Puerto Rico-Mayaguez, PR.
This work was supported by the National Science Foundation under the Engineering Research Centers Program, Award Number EEC – 9986821.
• Work in the estimation of the number of endmembers
• Research in the application of sparse positive matrix factorization as a way to improve the results in unmixing techniques.
[1] Y. .M. Masalmah, and M. Vélez-Reyes, “A full algorithm to compute the constrained positive matrix factorization and its application in unsupervised unmixing of hyperspectral.imagery.” In Proceedings of SPIE: Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XIV, Vol. 6966, April 2008.[2] Gruninger, J, A. J. Ratkowski and M. L. Hoke. “The Sequential Maximum Angle Convex Cone (SMACC) Endmember Model”. Proceedings SPIE, Algorithms for Multispectral and Hyper-spectral and Ultraspectral Imagery, Vol. 5425-1, Orlando FL, April, 2004[3]Lee, K., “A Subpixel Scale Target Detection Algorithm for Hyperspectral Imagery,” Ph.D. Dissertation, Rochester Institute of Technology, Center for Imaging Science, 2003.[4] Vélez-Reyes, Miguel and Rosario, Samuel. “Solving Adundance Estimation in Hyperspectral Unmixing as a Least Distance Problem”. IEEE. pp. 3276-3278.2004. [5] http://www.specmap.com/rem-hyper-aisa.html
From http:// www.ciudadseva.com/mapapr.htm
From http://www.elenas-vieques.com/roadmap.html
Vieques Island is located 22 miles east of Puerto Rico approximately . There is a need to build vegetation cover maps of the areas protected by US Fish and Wildlife Service.
The available data is from the AISA[5] sensor with spatial resolution of 3 meters and spectral information of 35 bands between 460 – 900 nanometers.
Water (cPMF)
0
0.5
1
Water (SMACC)
00.20.40.60.8
Water (Max D)
0
0.5
1
500 550 600 650 700 750 800 850 9000
100
200
300
400
500
600
700
800
900
1000Spectral Signatures Comparison of Water
Wavelength [Nanometers]
Am
plitu
de
Water (cPMF)
Water (SMACC)Water (Max D)
High trees (Flamboyan tree)(cPMF)
0
0.5
1
High trees (Flamboyan tree)(SMACC)
0
0.5
1
High trees (Flamboyan tree)(Max D)
0
0.5
1
500 550 600 650 700 750 800 850 9000
1000
2000
3000
4000
5000
6000Spectral Signature Comparison of Height Trees
Wavelength [Nanometers]
Am
plitu
de
Height trees(cPMF)
Height trees (SMACC)Height trees (Max D)HighHigh trees High trees
High trees
High trees
Black mangrove and palms (cPMF)
0
0.5
1
Black mangrove and palms
0
0.5
r (SMACC)
Black mangrove and palms (Max D)
0
0.5
500 550 600 650 700 750 800 850 9000
1000
2000
3000
4000
5000
6000Spectral Signatures Comparison of Black mangrove and palms
Wavelength [Nanometers]
Am
plitu
de
Black mangrove and palms (cPMF)
Black mangrove and palms (SMACC)Black mangrove and palms (Max D)
Bare Soil(cPMF)
0
0.5
1
Bare soil (SMACC)
0
0.5
1
Bare Soil (Max D)
0
0.5
1
500 550 600 650 700 750 800 850 9000
1000
2000
3000
4000
5000
6000Spectral Signatures Comparison of Bare Soil
Wavelength [Nanometers]
Am
plitu
de
Bare Soil(cPMF)
Bare Soil(SMACC)Bare Soil(Max D)
Red Mangrove(cPMF)
00.20.40.60.8
Shrubs (cPMF)
00.2
0.4
0.60.8
Red Mangrove - Shrubs (SMACC)
00.20.40.60.8
Red Mangrove - Shrubs (Max D)
0
0.5
1500 550 600 650 700 750 800 850 900
0
1000
2000
3000
4000
5000
6000Spectral Signatures Comparison of Red Mangrove and Shrubs
Wavelength [Nanometers]
Am
plitu
de
Red Mangrove(cPMF)
Shrubs(cPMF)Red Mangrove and Shrubs(SMACC)
Red Mangrove and Shrubs(Max D)
High trees (Flamboyan trees)Red MangroveWater Bare soil Black Mangrove and palm Shrubs
Water Results
Bare soil Results
High Trees Results
Black mangrove and palm Results
Red mangrove and Shrubs Results
[1] [2] [3]
[4]
This work is aimed at establishing the contribution of the work being done in unmixing in the R2C thrust. Procedures that lead to automated hyperspectral unmixing such as the cPMF are important for hyperspectral image exploitation.
Area of Interest