PolSAR image classification using generalized scattering models

In this paper, we present a new model-based method for polarimetric synthetic aperture radar (PolSAR) image classification. The conventional single- and double-bounce scattering models do not have contributions on T₃₃ element of coherency matrix. The T₃₃ element of coherency matrix accounts for the cross-polarization power. Surfaces having azimuth slopes and oriented man-made structures generate significant amount of cross-polarization power. In the proposed decomposition scheme, generalized single- and double-bounce scattering models are utilized to address this cross-polarization power. The proposed decomposition scheme is experimentally verified on Radarsat-2 San Francisco data. Experimental results are analyzed in terms of normalized means of scattering powers and percentage of negative power pixels which clearly indicate the effectiveness of the proposed decomposition scheme.