Evaluation of hybrid polarimetric decomposition techniques for winter crop discrimination

Abstract

In this paper we compare, using ISRO’s RISAT-1 FRS-1 mode Compact Polarimetric (CL-Pol) data, two widely used hybrid polarimetric decomposition techniques, m − δ and m − χ decompositions, with regard to classification accuracy for various agricultural crops of north and west India. We show that the classification based on the m − χ decomposition results in better crop separability in general. But the crop stage and existence of orientating structures in the crops affects the efficacy of decomposition; a fact vividly brought out in this paper. Theoretical insights into the effectiveness of these decomposition techniques for different crop geometry are brought forth. We also compare the classification accuracy subsequent to polarimetric speckle filtering vis-a-vis spatial multilooking (downsampling). We show that usage of an appropriate polarimetric filter tends to produce comparable accuracy for most of the agricultural classes, as that of multilook case, without degrading spatial resolution. This work showcases a custom implementation of Stokes parameter based decomposition as well as POLSAR filter based on refined Lee algorithm, written in C and tailored to RISAT-1.