Classification of Polarimetric SAR Images with Scattering Model and Wishart Classifier

This paper presents a new unsupervised classification of PolSAR images with scattering model and Wishart classifier. Firstly, a three-component scattering model with deorientation was applied to classify PolSAR images into three fundamental scattering categories and a mixed scattering category. Then, pixels in each kind of fundamental scattering categories were divided into more clusters based on the power of dominant scattering ones. After that, small clusters of each category were merged based on the Wishart distance between clusters. Finally, PolSAR data was classified iteratively with Wishart classifier to achieve unsupervised classification results. Real polarimetric datas collected with U.S. AIRSAR system are used to verify the new scheme, experimental results show the new algorithm improves classification accuracy and lowers volume scattering overestimation compared with traditional classification.