Improved SLIC superpixel generation algorithm and its application in polarimetric SAR images classification

Abstract

In recent years, more attention has been attracted on the classification of polarimetric SAR (PolSAR) images and a lot of methods have been proposed. With the resolution increasing, the pixel-based classification methods reveal insufficiency, therefore, this paper proposes an improved SLIC superpixel algorithm for PolSAR images classification. Firstly, effective polarization features, such as polarimetric scattering power of typical scattering mechanism based on target decomposition and spatial texture information based on statistical analysis, are extracted from PolSAR images and these features construct a feature vector to obtain a better description of PolSAR images. Then, the Euclidean distance of the feature vector is used to improve the SLIC algorithm to obtain superpixels segmentation and meantime reduce the execution time. In addition, a useful dissimilarity measurement is implemented to maintain the edges of different area in PolSAR image. At last, based on the superpixel segmentation result, a region-based classification using SVM is conducted. The proposed method is validated by EMISAR test PolSAR image and the experimental results confirm the performance and potential of the proposed method in PolSAR image interpretation.