Polarimetric SAR data classification with Freeman entropy and anisotropy analysis

The unsupervised classification of preserving polarimetric scattering characteristics is a classic classification method. But this method cannot classify the different objects with similar main scattering mechanism powers, especially for shadow, water and bare soil, which have very low backscattering powers. So the entropy and anisotropy parameters based on Freeman three-component decomposition is introduced, and applied into polarimetric SAR classification. Before applying the decomposition, a polarimetric orientation compensation (POC) procedure is performed for a better result. And then, the entropy Hf and anisotropy Af are calculated after Freeman decomposition. Through choosing appropriate values of Hf and Af, the shadow and water can be extracted out. The other pixels are then divided into three categories by their dominant scattering mechanisms. Each category is divided into 25~100 classes by the Hf~Af plane to preserve the purity of scattering characteristics, and merged into specified number of classes by Wishart distance measure. At last pixels in each category are iteratively classified by the Wishart classifier independently. A Radarsat-2 C band polarimetric SAR image was used to illustrate the effectiveness of the proposed method.