Superpixel-based classification of polarimetric synthetic aperture radar images

Abstract

Nowadays, polarimetric synthetic aperture radar (PolSAR) image classification is an important and widely studied topic. To overcome the limitations of pixel-based classification methods, we present, in this paper, a novel superpixel-based classification framework for PolSAR images. The framework takes the spatial relations between pixels into account and fully uses the statistical characteristics and contour information of PolSAR data. The framework is capable of integrating various inherent features of PolSAR data, improving classification accuracies, and making the results more understandable. Experiments on the AIRSAR data set show that the framework provides a promising solution for classifying PolSAR images.