Classification of Polarimetric SAR Data Using Spectral Graph Partitioning

Abstract

A new approach for classification of Polarimetric Synthetic Aperture Radar (POLSAR) data is proposed using segmentation that is formulated as a graph partitioning problem. This work is motivated by the fact that human experts are very good at visual interpretation and segmentation of POLSAR data, which is often challenging for automated analysis techniques. Spectral graph partitioning, a framework that has recently emerged in computer vision for solving grouping problems with perceptually plausible results, is used with modifications necessary to accommodate POLSAR data. Using the similarity of edge- aligned patch histograms and spatial proximity, classification performance that is superior to the Wishart classifier is achieved. This approach also provides a way to combine region-based and contour-based segmentation techniques, as it can accommodate different representations of polarimetric data as well as other data sources (e.g., optical imagery).