K-Wishart Classifier for PolSAR Data and Its Performance Evaluation

The complex K-Wishart distribution aims to describe the multilook covariance matrix and coherence matrix of polarimetric SAR (PolSAR) data by a statistical method to indicate the Non-Gaussian statistical characteristics of heterogeneous scenarios. Given the attractiveness of this approach for accurate description, it was applied to PolSAR classification. Two comparative unsupervised classification experiments using the Wishart and K-Wishart classifiers were designed for domestic airborne full-polarimetric SAR data from the Yigen test site in Inner Mongolia and the Zunhua test site in Hebei province. Preliminary results suggest that K-Wishart classifier is more suitable for extracting forests and buildings. The K-Wishart classifier performance was also evaluated in terms of overall accuracy and stability.