ALOS 2008 Symposium, 3-7 November Rhodes, Greece Comparison of ALOS-PALSAR and TerraSAR-X Data in terms of Detecting Settlements First Results Thomas Esch*, Achim Roth*, Michael Thiel, Michael Schmidt*, Stefan Dech* * German Aerospace Center (DLR), German Remote Sensing Data Center (DFD), Oberpfaffenhofen University of Würzburg, Department of Remote Sensing in cooperation with DLR German Aerospace Center (DLR) German Remote Sensing Data Center (DFD)
Outline Research Objectives Data and Software Methodology Results Conclusions Perspectives Folie 2
Research objectives Demonstrate and compare the potential of co-polarized TerraSAR- X and ALOS-PALSAR images for a delineation of urban footprints Transfer of existing method developed for TerraSAR-X data to ALOS L-band images Comparative assessment of the results Investigating the robustness of the procedure Attributing differences due to system and imaging parameters Folie 3
Data and software Data Rhine-Neckar Eiderstedt Istanbul TSX ALOS TSX ALOS TSX ALOS Date 06-07-2007 23-07-2007 01-09-2008 07-04-2008 17-07-2007 07-05-2007 Product type SM-GEC FM-MGD SM-MGD FM-MGD SM-MGD FM-MGD Coverage [km 2 ] 2,100 4,700 1,700 5,500 1.570 3,600 Polarization HH HH VV HH HH HH Geom. Resolution [m] 2.75 6.25 1.25 6.25 1.75 12.5 Incidence angle [deg.] 35.3 34.3 28.8 34.3 41.1 21.5 Software: IDL/ENVI, Definiens Developer Folie 4
Methodology Schematic view on image pre-processing and analysis procedure Folie 5
Methodology Pre-processing: Speckle analysis and suppression Novel approach: Fully automated analysis of scene-specific level of speckle noise Derivation of textural information based on speckle analysis (speckle divergence) Spatially adaptive and radiometrically selective reduction of SAR speckle Folie 6
Methodology Pre-processing: Speckle analysis Local deviation from scene-specific level of speckle (TSX) Topographic map Folie 7
Methodology Pre-processing: Speckle analysis TSX intensity ALOS intensity Speckle divergence Speckle divergence Folie 8
Methodology Pre-processing: Speckle suppression Filtered TerraSAR-X TSX intensity image Folie 9
Methodology Pre-processing: Speckle suppression Filtered ALOS intensity image Folie 10
Methodology Image analysis Folie 11
Methodology Image analysis: Segmentation (coarse level L1) Identification of potential built-up areas Filtered intensity image Coarse object level Folie 12
Methodology Image analysis: Pre-classifcation (coarse level L1) Identification of potential built-up areas High speckle divergence High difference between brightest to darkest pixel Potential built-up area Folie 13
Methodology Image analysis: Segmentation (fine level L0) Object basis for detailed image classification Filtered intensity image Fine object level Folie 14
Methodology Image Classification (L0): Distinct scatterers Conditions: - Very high intensity - Very high speckle divergence Folie 15
Methodology Image Classification (L0): Urban scatterers Conditions: - High intensity - High speckle divergence - Numerous distinct and urban scatterers in surrounding area Folie 16
Methodology Image Classification (L0): Urban footprint Conditions: - Super-object with high intensity - Super-object with high speckle divergence - Significant number of distinct and urban scatterers in surrounding area - Medium speckle divergence Folie 17
Results Rhine-Neckar Overall accuracy: 77% Overall accuracy: 95% Urban footprints derived from ALOS Urban footprints derived from TerraSAR-X Folie 18
Results Eiderstedt Overall accuracy: 81% Overall accuracy: 94% Urban footprints derived from ALOS Urban footprints derived from TerraSAR-X Folie 19
Results Istanbul Overall accuracy: 81% Overall accuracy: 96% Urban footprints derived from ALOS Urban footprints derived from TerraSAR-X Folie 20
Results Challenges and differences ALOS TerraSAR-X ALOS TerraSAR-X Folie 21
Conclusions One single-pol SAR scene is sufficient for settlement detection Methodology could be transferred from TerraSAR-X to ALOS-PALSAR Speckle divergence is essential parameter to keep rule base simple Urban footprints derived from ALOS L-band data show decrease in accuracy of 15 20 % Lower geometric resolution Considerable alteration in the appearance of specific objects and land cover types compared to X-band images Certain land cover types or areas are still challenging to both systems: Allotments or intensively greened residential areas Mountainous regions Folie 22
Perspectives Improvement/assessment of presented method Test sites located in other natural/cultural environments Influence of imaging parameters Frequency Spatial resolution Viewing geometry Further developments Change detection Land cover classification (water, settlement, woodland, open land) based on speckle analysis Folie 23
Thank you for your attention! Contact: Dr. Thomas Esch Thomas.Esch@dlr.de Achim Roth Achim.Roth@dlr.de Michael Thiel Michael.Thiel@uni-wuerzburg.de Folie 24