Comparison of polarimetric SAR features for terrain classification using incremental training

Abstract

In this study, the most commonly used polarimetric SAR features including the complete coherency (or covariance) matrix information, features obtained from several coherent and incoherent target decompositions, the backscattering power and the visual texture features are compared in terms of their classification performance of different terrain classes. For pattern recognition, two powerful machine learning techniques, Collective Network of Binary Classifier (CNBC) with incremental training capability and Support Vector Machines (SVM) are employed. Each feature has its own strength and weaknesses for discriminating different SAR class types and this study aims to investigate them through incremental feature based training of both classifiers and compare the results of the experiments performed using the fully polarimetric San Francisco Bay and Flevoland datasets.