Classification Using Optimal Polarimetric Parameters for Compact Polarimetric Data

Prior to classification, Polarimetric Synthetic Aperture Radar (PolSAR) research emphasizes the selection of polarimetric parameters for each land cover class. Polarimetric parameters are crucial to the identification of a target as each parameter has varied capability for target determination. By selecting optimal parameters, classification process can be improved. It is suggested that optimal parameters for each class be selected, to enhance classification accuracy. In this paper, a separability analysis is conducted to determine the optimal polarimetric parameters for distinguishing between various categories of land cover. In the case of hybrid polarimetric data, although only two scattering elements (RH and RV) are used, twenty polarimetric parameters are determined. Jeffries-Matusita distance is used to identify the most separable bands for each land cover type. Selected bands are then used for classification, and visual analysis reveals that the classification precision computed using selected bands is high.