Model-based target classification using polarimetric similarity with coherency matrix elements

Abstract

Polarimetric similarity is a parameter for measuring the similarity between two scattering mechanisms. In this paper, we propose a novel model-based target classification technique using a compensated polarimetric similarity parameter between two coherency matrices. In general, the ensemble average coherency matrix elements have magnitude imbalance, thus the contribution degree to the polarimetric similarity differs for each element. We illustrate how to compensate the contribution degree, and then the proposed method is tested on L-band fully polarimetric ALOS-2/PALSAR-2 data sets by using 4 theoretical scattering models (surface scattering, double-bounce scattering, volume scattering, and 22.5° oriented dihedral scattering). The classification results show that the new compensation scheme serves to better classification.