Semi-automatic Polarimetric SAR Image Classification by MD PSO Based Dynamic Clustering

Abstract

In this study, a new systematic approach for semi-automatic classification of polarimetric synthetic aperture radar (PoISAR) image is proposed. The feature extraction block utilizes traditionally used SAR features including the complete coherency (or covariance) matrix information, features derived from various target decomposition theorems, the backscattering power and the selected texture features from gray-level cooccurrence matrix (GLCM). Classification of the information in multi-dimensional PoISAR data space by dynamic clustering is addressed as an optimization problem and recently proposed multi-dimensional particle swarm optimization (MD PSO) technique is applied to find optimal clusters in a given input data space, distance metric and a proper validity index function. An experimental study is performed using the fully polarimetric San Francisco Bay AIRSAR dataset to analyze and compare the results of classification with the state of the art techniques.