Unsupervised Classification of Land-Coverage Using Polarimetric SAR images

Abstract— Widely used unsupervised classification method H/A/alpha classification, explores the scattering information of land-coverage data, but performs poorly on the decision boundary. Maximum Likelihood Supervised Classification using Wishart distribution, based on the statistic properties requiring picking up training set manually from large SAR image, can’t be automated. In this project, I propose a hybrid classification scheme combining both methods, exploit the scattering mechanism of targets and is also automated.