A knowledge-based approach for discriminating multi-crop scenarios using multi-temporal polarimetric SAR parameters

In this article, we evaluate a series of POLSAR (Polarimetric Synthetic Aperture Radar) parameters and devise a robust, multi-date, and hierarchical decision tree algorithm for crop discrimination. The study area is a farmland in North India having relatively large patches of winter crops with medium heterogeneity. Ten POLSAR parameters are evaluated for this work including polarimetric entropy (), polarimetric alpha angle (), Yamaguchi decomposition components, radar vegetation index (RVI), volume scattering index, canopy scattering index, biomass index, and pedestal height (PH). We choose polarimetric RVI, polarimetric entropy, and polarimetric alpha angle parameter, after sensitivity analysis, to be incorporated into building the multi-temporal POLSAR decision tree. The proposed algorithm accepts, as input, a precisely co-registered multi-date stack of fully POLSAR imagery. The algorithm makes use of the temporal profiles of the selected POLSAR parameters in achieving the crop discrimination. We provide a quantitative assessment of classification accuracy of various classes based on extensive ground truth data. The multi-temporal algorithm is compared with the well-established, single-date, supervised Wishart classifier. Statistically significant improvement of accuracies is observed across various classes as compared to single-date methodology. Our study suggests that whenever ground truth data are extensively available, a supervised classifier based on carefully chosen multi-temporal POLSAR parameters yield very compelling crop discrimination capabilities.