wAxb+ - CenSSIS
Graduate Student: Maria Constanza Torres-Madronero, [email protected] Dr. Miguel Velez-Reyes, [email protected], Dr. Skip Van Bloem, [email protected] Laboratory for Applied Remote Sensing and Image Processing, University of Puerto Rico at Mayagüez, P. O. Box 9048, Mayagüez, Puerto Rico 00681-9048
ABSTRACT Temporal analysis of remote sensing imagery allows the identification of environmental changes and seasonal behavior for biodiversity studies. This project is looking at the application of unmixing techniques for temporal analysis of hyperspectral imagery. Changes in endmembers and their abundances will be used to estimate seasonal changes in land cover and biodiversity. In this initial work, analysis of Hyperion imagery taken over the Guanica Dry forest in southwestern Puerto Rico during December 2004 is presented. Here we study how sensor configuration (e.g. sensor look angle) and scene features (e.g. Topography and atmospheric conditions) affect the results of the unmixing process METHODOLOGY
STATE OF ART HYPERSPECTRAL REMOTE SENSOR
ENDMEMBERS FROM HYPERION IMAGE 10 DEC 2004
1
1 end1 end2 end3 end4 end5 end6 end7
0.9
xˆ arg min Ax b
0.8 0.7
subjet to x 0 and 1 x 1 T
AMPLITUDE
SMACC
Endmember Estimation (i.e SMACC [2])
Unmixing Model [1]
0.6
10 Dec 26 Dec
0.9 0.8
Correlation Coef.
NNSTO
0.5
0.7 0.6
0.4
0.5
0.3
0.4
0.2
0.3 0.1
b Ax w
10 DEC 2004
0.2 600
800
1000
ENDMEMBERS FROM HYPERION IMAGE 10 DEC 2004 1
0.8 0.7 0.6
1800
2000
2200
0.8 0.7
0.3 0.2
AMPLITUDE
Abundance Estimation (i.e NNSTO [3])
0.1
800
1000
1200 1400 1600 WAVELENGTH
1800
2000
2200
0.1
2400
0 400
end1 end2 end3 end4 end5 end6 end7
0.9
0.4
600
End1: 10 DEC 2004
2400
1
0.5
0 400
1200 1400 1600 WAVELENGTH
ENDMEMBER FROM HYPERION IMAGE 17 DEC 2004
end1 end2 end3 end4 end5 end6 end7
0.9
AMPLITUDE
Applications • Target detection • Land cover classification • Environment monitoring • Military applications
0 400
0.6
600
800
1000
1200
1400
1600
1800
2200
2400
RHO = 0.9739 Guanica Land Cover Map
End1
0.5 0.4
Exposure Soil
0.3 0.2 0.1
Endmember Matrix
0 400
26 DEC 2004
.
2000
STUDY AREA
1000
1200 1400 1600 WAVELENGTH
1800
2000
2200
End1: 26 DEC 2004
2400
Endmember Correlation and Identification
Abundance Estimation
GORDON CenSSIS VALUE ADDED
Lowland dry limestone woodland and shrubland
Guanica Dry Forest is in the Southwest of Puerto Rico. It was declared as Biosphere Reserve in 1981 by UNESCO. The Guanica Dry Forest ecosystem includes several zones of grassland, shrubs, cactus and mangrove forest.
800
Endmember Extraction
Multi-temporal HSI Young secondary lowland dry limestone semideciduous forest
600
RESEARCH TO REALITY
Temporal analysis from HSI data provides support for CenSSIS Researchers and Students in R2C .This techniques can be used for management and studies of terrestrial and coastal ecosystems.
Dry cactus grassland and shrubland Mature secondary lowland dry limestone evergreen forest Mangrove forest and shrubland
Lowland dry alluvial shrubland and woodland Salt and mudflats
Low-density urban development
Land Cover Classification Map [4]
Validation
Temporal analysis from hyperspectral imagery allows the change detection of land cover and the biodiversity seasonally assessment. These results can be part of biodiversity conservation programs and environmental studies. Study area of our research is Guanica Dry Forest, a Biosphere Reserve of UNESCO from 1981. Results of this research can help to understand the biodiversity dynamic of the region. We expect that our results will contribute to the conservation of this unique place in Puerto Rico.
UNMIXING OF MULTI-TEMPORAL HYPERSPECTRAL IMAGERY HYPERION IMAGE FOR DECEMBER 2004 [5] 1
1
1
0.9
Endmember 1
Endmember 2 Endmember 3
Endmember 4
Endmember 5
Endmember 6 Endmember 7
0.8
0.7
3 Dec
0.7
0.6
0.6
0.5
0.5
3 Dec Time: 14:32 SENSOR_LOOK_ANGLE = -12.738
10 Dec
Time: 14:38 SENSOR_LOOK_ANGLE = 0.0349
0.4
0.3
0.3
0.2
0.2
0.1 0 400
0.1
600
800
1000
1200
1400
1600
1800
2000
2200
2400
0 400
800
1000
1200
Time: 14:44 SENSOR_LOOK_ANGLE = 12.614
27 Dec Time: 14:40 SENSOR_LOOK_ANGLE = 0.0623
Hyperspectral image captured by HYPERION sensor in four different days at December of 2004. HSI consists of 220 bands from 0.4 to 2.5µm, with a spatial resolution of 30 meters. In this experiment, only 145 bands are used and a mask was applied to remove the water bodies.
1
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
600
800
1000
1200
1400
1600
1800
2000
2200
0.9 0.8
0.7
0.7
0.6
0.6
0.6
0.6
0.5
0.5
0.5
0.5
0.4
0.4
0.4
0.4
0.3
0.3
0.3
0.3
0.2
0.2
0.1
0.1
600
800
1000
1200
1400
1600
1800
2000
2200
2400
0 400
600
800
1000
1200
1400
1600
1800
2000
2200
RHO
3 Dec
10 Dec
17 Dec
26 Dec
3 Dec
1
0.9840
0.9847
0.9822
10 Dec
0.9840
1
0.9834
0.9920
17 Dec
0.9847 0.9822
0.9834 0.9920
0.1 0 400
2400
1 0.9790
© 2010 Google: Image U.S. Geological Survey
0.2
600
800
1000
1200
1400
1600
1800
2000
2200
RHO
3 Dec
10 Dec
26 Dec
3 Dec
1
0.8454
0.8265
10 Dec
0.8454
1
0.8024
26 Dec
0.8265
0.8024
1
RHO = 0.9178
0.9
3 Dec 17 Dec
RHO = 0.9774
0.8 0.7 0.6 0.5 0.4 0.3 0.2
0.2
0.1
0.1 0 400
2400
10 Dec 26 Dec
600
800
1000
0.9790 1
1200
1400
1600
1800
2000
2200
2400
© 2010 Google: Image U.S. Geological Survey
2400
0 400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
© 2010 Google: Image U.S. Geological Survey
10 Dec
1
0 400
1400
0.8
© 2010 Google: Image U.S. Geological Survey
1
1 3 Dec 10 Dec 26 Dec
0.9
0.7
26 Dec 600
3 Dec
17 Dec
0.8
© 2010 Google: Image U.S. Geological Survey
10 Dec
1 3 Dec 10 Dec 17 Dec 26 Dec
0.9
0.7
0 400
0.4
10 Dec 26 Dec
RHO = 0.9739
0.8
0.9
0.8
1
0.9
1600
17 Dec
1800
2000
2200
2400
0 400
17 Dec 3 Dec
10 Dec
3 Dec
10 Dec
10 Dec
17 Dec
26 Dec
800
1000
1200
1400
1600
1800
2000
26 Dec
2200
3 Dec
26 Dec 26 Dec
0 600
10 Dec
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
2400
26 Dec
Abundance map for HyPERION image of GuanicA Dry Forest obtained using NNSTO.
26 Dec Guanica Land Cover 2006 [4]
20
Guanica Land Cover 2006 [4]
17 Dec
40 60
ACKNOWLEDGMENTS
FUTURE WORK
REFERENCES
• Temporal analysis of images captured in a same month different years. • Process images for a full year. • Understand view angle dependence. • Develop biodiversity indices from unmixing results.
[1] N.Kashava; J.F.Mustrad. “spectral unmixing”. IEEE Signal Processing Magazine, 5:44–57, 2002. [2] J. Gruninger, A. J. Ratkowski, and M. L. Hoke. “The Sequencial Maximum Angle Convec Cone (SMACC) Endmember Model”. In SPIE Proc. 5425-1, April 2004. [3] S.Rosario. Iterative Algorithms for Abundance Estimation on Unmixing of Hyperspectral Imagery. Master’s thesis, University of Puerto Rico, 2004. [4] W. A. Gould. “The Puerto Rico GAP Analysis Report – Final Report”. July 2007. [5] USGS data available onlineL http://eo1.usgs.gov/
80 100
This work was primarily supported by the NASA EPSCoR Program under grant NNX09AV03A and used facilities supported by the NSF Engineering Research Centers Program under grant ECC-9986821. Maria C. Torres-Madronero was also supported by a Gordon CenSSIS Bridge Scholarship.
120 140 160
50
100
150
200
250
ABSTRACT Temporal analysis of remote sensing imagery allows the identification of environmental changes and seasonal behavior for biodiversity studies. This project is looking at the application of unmixing techniques for temporal analysis of hyperspectral imagery. Changes in endmembers and their abundances will be used to estimate seasonal changes in land cover and biodiversity. In this initial work, analysis of Hyperion imagery taken over the Guanica Dry forest in southwestern Puerto Rico during December 2004 is presented. Here we study how sensor configuration (e.g. sensor look angle) and scene features (e.g. Topography and atmospheric conditions) affect the results of the unmixing process METHODOLOGY
STATE OF ART HYPERSPECTRAL REMOTE SENSOR
ENDMEMBERS FROM HYPERION IMAGE 10 DEC 2004
1
1 end1 end2 end3 end4 end5 end6 end7
0.9
xˆ arg min Ax b
0.8 0.7
subjet to x 0 and 1 x 1 T
AMPLITUDE
SMACC
Endmember Estimation (i.e SMACC [2])
Unmixing Model [1]
0.6
10 Dec 26 Dec
0.9 0.8
Correlation Coef.
NNSTO
0.5
0.7 0.6
0.4
0.5
0.3
0.4
0.2
0.3 0.1
b Ax w
10 DEC 2004
0.2 600
800
1000
ENDMEMBERS FROM HYPERION IMAGE 10 DEC 2004 1
0.8 0.7 0.6
1800
2000
2200
0.8 0.7
0.3 0.2
AMPLITUDE
Abundance Estimation (i.e NNSTO [3])
0.1
800
1000
1200 1400 1600 WAVELENGTH
1800
2000
2200
0.1
2400
0 400
end1 end2 end3 end4 end5 end6 end7
0.9
0.4
600
End1: 10 DEC 2004
2400
1
0.5
0 400
1200 1400 1600 WAVELENGTH
ENDMEMBER FROM HYPERION IMAGE 17 DEC 2004
end1 end2 end3 end4 end5 end6 end7
0.9
AMPLITUDE
Applications • Target detection • Land cover classification • Environment monitoring • Military applications
0 400
0.6
600
800
1000
1200
1400
1600
1800
2200
2400
RHO = 0.9739 Guanica Land Cover Map
End1
0.5 0.4
Exposure Soil
0.3 0.2 0.1
Endmember Matrix
0 400
26 DEC 2004
.
2000
STUDY AREA
1000
1200 1400 1600 WAVELENGTH
1800
2000
2200
End1: 26 DEC 2004
2400
Endmember Correlation and Identification
Abundance Estimation
GORDON CenSSIS VALUE ADDED
Lowland dry limestone woodland and shrubland
Guanica Dry Forest is in the Southwest of Puerto Rico. It was declared as Biosphere Reserve in 1981 by UNESCO. The Guanica Dry Forest ecosystem includes several zones of grassland, shrubs, cactus and mangrove forest.
800
Endmember Extraction
Multi-temporal HSI Young secondary lowland dry limestone semideciduous forest
600
RESEARCH TO REALITY
Temporal analysis from HSI data provides support for CenSSIS Researchers and Students in R2C .This techniques can be used for management and studies of terrestrial and coastal ecosystems.
Dry cactus grassland and shrubland Mature secondary lowland dry limestone evergreen forest Mangrove forest and shrubland
Lowland dry alluvial shrubland and woodland Salt and mudflats
Low-density urban development
Land Cover Classification Map [4]
Validation
Temporal analysis from hyperspectral imagery allows the change detection of land cover and the biodiversity seasonally assessment. These results can be part of biodiversity conservation programs and environmental studies. Study area of our research is Guanica Dry Forest, a Biosphere Reserve of UNESCO from 1981. Results of this research can help to understand the biodiversity dynamic of the region. We expect that our results will contribute to the conservation of this unique place in Puerto Rico.
UNMIXING OF MULTI-TEMPORAL HYPERSPECTRAL IMAGERY HYPERION IMAGE FOR DECEMBER 2004 [5] 1
1
1
0.9
Endmember 1
Endmember 2 Endmember 3
Endmember 4
Endmember 5
Endmember 6 Endmember 7
0.8
0.7
3 Dec
0.7
0.6
0.6
0.5
0.5
3 Dec Time: 14:32 SENSOR_LOOK_ANGLE = -12.738
10 Dec
Time: 14:38 SENSOR_LOOK_ANGLE = 0.0349
0.4
0.3
0.3
0.2
0.2
0.1 0 400
0.1
600
800
1000
1200
1400
1600
1800
2000
2200
2400
0 400
800
1000
1200
Time: 14:44 SENSOR_LOOK_ANGLE = 12.614
27 Dec Time: 14:40 SENSOR_LOOK_ANGLE = 0.0623
Hyperspectral image captured by HYPERION sensor in four different days at December of 2004. HSI consists of 220 bands from 0.4 to 2.5µm, with a spatial resolution of 30 meters. In this experiment, only 145 bands are used and a mask was applied to remove the water bodies.
1
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
600
800
1000
1200
1400
1600
1800
2000
2200
0.9 0.8
0.7
0.7
0.6
0.6
0.6
0.6
0.5
0.5
0.5
0.5
0.4
0.4
0.4
0.4
0.3
0.3
0.3
0.3
0.2
0.2
0.1
0.1
600
800
1000
1200
1400
1600
1800
2000
2200
2400
0 400
600
800
1000
1200
1400
1600
1800
2000
2200
RHO
3 Dec
10 Dec
17 Dec
26 Dec
3 Dec
1
0.9840
0.9847
0.9822
10 Dec
0.9840
1
0.9834
0.9920
17 Dec
0.9847 0.9822
0.9834 0.9920
0.1 0 400
2400
1 0.9790
© 2010 Google: Image U.S. Geological Survey
0.2
600
800
1000
1200
1400
1600
1800
2000
2200
RHO
3 Dec
10 Dec
26 Dec
3 Dec
1
0.8454
0.8265
10 Dec
0.8454
1
0.8024
26 Dec
0.8265
0.8024
1
RHO = 0.9178
0.9
3 Dec 17 Dec
RHO = 0.9774
0.8 0.7 0.6 0.5 0.4 0.3 0.2
0.2
0.1
0.1 0 400
2400
10 Dec 26 Dec
600
800
1000
0.9790 1
1200
1400
1600
1800
2000
2200
2400
© 2010 Google: Image U.S. Geological Survey
2400
0 400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
© 2010 Google: Image U.S. Geological Survey
10 Dec
1
0 400
1400
0.8
© 2010 Google: Image U.S. Geological Survey
1
1 3 Dec 10 Dec 26 Dec
0.9
0.7
26 Dec 600
3 Dec
17 Dec
0.8
© 2010 Google: Image U.S. Geological Survey
10 Dec
1 3 Dec 10 Dec 17 Dec 26 Dec
0.9
0.7
0 400
0.4
10 Dec 26 Dec
RHO = 0.9739
0.8
0.9
0.8
1
0.9
1600
17 Dec
1800
2000
2200
2400
0 400
17 Dec 3 Dec
10 Dec
3 Dec
10 Dec
10 Dec
17 Dec
26 Dec
800
1000
1200
1400
1600
1800
2000
26 Dec
2200
3 Dec
26 Dec 26 Dec
0 600
10 Dec
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
2400
26 Dec
Abundance map for HyPERION image of GuanicA Dry Forest obtained using NNSTO.
26 Dec Guanica Land Cover 2006 [4]
20
Guanica Land Cover 2006 [4]
17 Dec
40 60
ACKNOWLEDGMENTS
FUTURE WORK
REFERENCES
• Temporal analysis of images captured in a same month different years. • Process images for a full year. • Understand view angle dependence. • Develop biodiversity indices from unmixing results.
[1] N.Kashava; J.F.Mustrad. “spectral unmixing”. IEEE Signal Processing Magazine, 5:44–57, 2002. [2] J. Gruninger, A. J. Ratkowski, and M. L. Hoke. “The Sequencial Maximum Angle Convec Cone (SMACC) Endmember Model”. In SPIE Proc. 5425-1, April 2004. [3] S.Rosario. Iterative Algorithms for Abundance Estimation on Unmixing of Hyperspectral Imagery. Master’s thesis, University of Puerto Rico, 2004. [4] W. A. Gould. “The Puerto Rico GAP Analysis Report – Final Report”. July 2007. [5] USGS data available onlineL http://eo1.usgs.gov/
80 100
This work was primarily supported by the NASA EPSCoR Program under grant NNX09AV03A and used facilities supported by the NSF Engineering Research Centers Program under grant ECC-9986821. Maria C. Torres-Madronero was also supported by a Gordon CenSSIS Bridge Scholarship.
120 140 160
50
100
150
200
250
