3_Terrain catergorization for single Pol.ppt
- 1. Terrain Categorization Based on Scattering Mechanisms for Single-Pol High-Resolution TerraSAR-X images IGARSS 2011 – Vancouver, Canada July 24 - July 29 Jong-Sen Lee, Thomas Ainsworth, Naval Research laboratory Washington DC 20375, USA Kun-Shan Chen CSRSR, National Central University, Taiwan Irena Hajnsek German Aerospace Research Center (DLR), Germany
- 2. Introduction TerraSAR-X has provided high-resolution data in single-pol, dual-pol Advantages of single-pol Large swath for coverage Data handling High-resolution (TerraSAR-X spotlight-mode) Deficiencies of single-pol Separating scattering mechanisms – Unlike PolSAR Terrain and land-use classification Terrain categorization Color single-pol similar to Pauli decomposition of PolSAR
- 3. PolSAR – Pauli Decomposition Quad-Pol SAR Pauli Vector Volume (Canopy) Double Bounce Rough Surface
- 4. Quad-Pol Capability (L-band JPL/AirSAR) Pauli |HH-VV|, |HV |, |HH+VV| 4 th Iteration (15 classes) J.S. Lee, M.R. Grunes, E. Pottier, L. Ferro-Famil, “Unsupervised terrain classification preserving scattering characteristics,” IEEE Transactions on Geoscience and Remote Sensing,vol. 42, no.4, pp. 722-731, April, 2004.
- 5. PolSAR – Pauli Decomposition Single-Pol SAR – Only one observable in |HH| or |VV| Surface: Lower returns Volume: Middle returns Double bounce: Higher returns Surface Volume DB Ocean/water Bare Ground Trees Buildings Difficulties: Speckle Scattering categories overlap Retaining resolution
- 6. What to be attempted for single-pol? Color-coding each pixel Ocean and water surface, bare surface Forest, Trees Buildings, hard targets (along track direction) A Challenge!
- 7. The Procedure Speckle filtering by applying the Improved Sigma Filter Unsupervised classification into classes Scattering category determination Color each category accordingly Split color coded map into Hue Saturation Intensity The Intensity image is replaced with The original amplitude image (unfiltered) – retain resolution Combine (Hue, Saturation, Amplitude ) and produce the color-coded categorized image
- 8. The Illustrated Procedure TerraSAR-X Spotlight Mode HH-Polarization 4578 x 8448 pixels, 2 pixels averaged (azimuth) Additional 2x2 average to reduce 2289 x 2112
- 9. The Illustrated Procedure Apply the improved sigma filter twice (over- filtered) for speckle reduction
- 10. The Illustrated Procedure Unsupervised classification into 8 to 16 classes Divided all pixels into 8 classes Apply Wishart classification for 7 iterations
- 11. The Illustrated Procedure Color rendering – important step, decision by calibrated intensity and manual decision may be needed. Surface has divided into two
- 12. The Illustrated Procedure To preserve spatial resolution, split colored map into ( H, S, I ) Colored map Hue Saturation Intensity
- 13. The Illustrated Procedure Combine Hue and Saturation with the original unfiltered image to produce final results. Hue Saturation Original (unfiltered)
- 14. The Illustrated Procedure Combine Hue and Saturation with the original unfiltered image to produce final results. The original amplitude Terrain categorized color image
- 15. Zoom-in Comparison The original Color-coded original
- 16. Comparison with Google Earth Different image acquisition dates Google Earth image Final results
- 17. Comparison with PolSAR Different image acquisition dates PolSAR Pauli Single-pol - Final results
- 18. Another Example TerraSAR-X HH-Polarization – Taipei City, Taiwan 18772 x 27721 pixels, cropped out 6336 x 6000 pixels Acquisition mode: strip_007, 3 meter resolution The original amplitude Terrain categorized color image
- 19. Another Example TerraSAR-X HH-Polarization – Taipei City, Taiwan The original amplitude Terrain categorized color image
- 20. Another Example TerraSAR-X HH-Polarization – Taipei City, Taiwan The original amplitude Terrain categorized color image
- 21. Remarks and Discussion Easier to separate surface from volume Speckle index (texture) can be applied Difficult to separate volume from DB Buildings not aligned in az. direction could be miss-assigned to volume due to orientation angle effect Rough ocean surface could have higher returns than land surface (surface divided into two) Simple manual procedure was implemented Zeros (0.5%) in TerraSAR-X data, due to thresholding applied, requires special attention Affecting classification and speckle filtering Manual color assignment may be needed HH-pol is better than VV-pol
- 22. Conclusion Terrain categorization to color the single-pol SAR data based on Pauli decomposition of PolSAR Large dimensional data demand efficient algorithm Preserve spatial resolution for high-resolution data The algorithm is not perfect – errors are expected Simple manual procedure was implemented Terrain categorization for dual-pol - completed HV-pol enhance volume scattering Single-pol cannot replace quad-pol PolSAR can classify much more accurately Scattering mechanisms can be readily available Other PolSAR applications – geophysical parameter, orientation angle, target characterization, Pol-InSAR, etc.
- 23. Special procedure for Rough Ocean Surface