Spectral Clustering ensemble for polarimetric sar classification with Wishart-derived distance and Polarimetric similarity

In this paper, a new method using spectral clustering ensemble for PolSAR classification is proposed. Diverse basic classifications are performed on PolSAR data to apply Nyström approximation method of Spectral Clustering to PolSAR classification and improve its robustness to scaling parameter. Then, all the basic classifications are assembled to obtain the final classification of PolSAR data. During the process of Spectral Clustering, Wishart-derived distance measure and Polarimetric similarity are combined together to consider space and detail relations between pairwise pixels. The Wishart classifier, which is designed for PolSAR data, is employed to perform classification on PolSAR images and achieve accurate results. Experiments are provided to verify the effectiveness of the proposed method. The simulation results illustrate that the proposed method outperforms the comparisons without employing ensemble strategy and those with only one simple similarity measure.