Superpixel-Based Unsupervised Classification of PolSAR Imagery Using Wishart Mixture Models and Spectral Clustering

摘要

Unsupervised classification of polarimetric synthetic aperture radar (PolSAR) imagery is an essential step in SAR image interpretation. In this paper, we propose a framework for superpixel-based unsupervised classification of PolSAR imagery. Firstly, the SLIC super-pixel algorithm is adapted for generating compactness local regions. Secondly, Wishart Mixture Models (WMM) are learned to model each local region and two analytic information-theoretic divergences are employed for computing the (dis)similarities of region pairs. Finally, the classification results are obtained by using the spectral clustering approach. The experimental results on different SAR data sets show the effectiveness of our method.