Unsupervised Classification Methods and Experimental Research of Dual-frequency Fully Polarimetric SAR Images
双波段全极化SAR图像非监督分类方法及实验研究

In this paper, initial assumption of SAR pixel distribution is derived from H/a classifier. Then a Maximum Likelihood (ML) method is introduced to improve the classifi-cation.. The backscattering properties of a natural medium, that varies with the observation frequency, dual-frequency SAR images are combined to get further improved classification. Speckle in SAR images will disturb classification accuracy. Vector filter of speckle is used to dual-frequency images before classification. Experiments are done on data got by NASA/JPL lab near Tien Mountains, and pseudo-colored classification results of both single and dual frequency POLSAR image are submitted. Results show that filtered dual-frequency fully polarimetric SAR data obtain the best classification result.