Classification of building area using azimuth angle and density indices derived from polarimetric SAR

Urban extraction is one of the most anticipated applications of remote sensing, but automatic extraction has been challenging. Synthetic aperture radar (SAR) applications are particularly problematic because the scattering in urban areas is sensitive to the spatial arrangement of buildings, which prevents extraction. Spaceborne polarimetric synthetic aperture radar (POLSAR), an advanced approach to SAR, has been available since ALOS/PALSAR was launched in 2006. Several indicators derived from POLSAR data have been developed to classify land cover, and some of them have been utilized to extract the geometric features of a target. So, as a first step, we measured the backscattering from concrete blocks arranged with different azimuth angles and distances in an anechoic radio wave chamber. Through this experiment, we found the interrelation between spatial arrangement and various polarimetric indices, and demonstrated that Polarization Orientation Angle (POA) is a good detector for the azimuth angle of man-made structures. We also found that entropy has the highest correlation with building density. Then we classified urban areas in satellite polarimetric data using POA and entropy. A comparison with Quickbird image indicated that POA is an effective indicator to extract non-perpendicular buildings and that there are some areas where entropy distinguishes differences in building density.