Unsupervised Classification of Polarimetric SAR Image Using Deorientation Theory and Complex Wishart Distribution
基于去取向理论的全极化SAR图像模糊非监督聚类

Scatter targets of complex terrain surfaces with random orientation product random fluctuating echoes. This leads to a confused classification by directly using target decomposition on full polarimetric SAR (PolSAR) image. To solve this problem, a new unsupervised classification method is proposed in this paper. Firstly, the target vector is transformed to the state with minimization of cross-polarization (min-x-pol); then the parameters u/v/H are used to characterized scattering mechanism, and the fuzzy membership is adopted instead of "hard" division of parameter plan; finally, characterizing the coherency matrix as multivariable complex Wishart distribution, the polarimetric SAR image is classified based on Bayes maximum likelihood criteria. Experiment is performed on a L-band NASA/JPL SIR-C polarimetric SAR image over Danshui town, Guangdong, P.R. China. Furthermore, the movements of the clustering centers are discussed. Compared with the k-mean like method based on , the results show that the proposed method provides a significant performance improvement in classification result and the associated scattering mechanism of class is more accurate. The classification result is beneficial for automatic recognition of terrain type.