基于非高斯分布的全极化SAR数据无监督分类

摘要

In this paper, an unsupervised algorithm is proposed to classify the data of fully polarimetric synthetic aperture Radar(PolSAR).The proposed method combines the Freeman-Durden with the scattering model based development of the decomposition algorithm and the K-Wishart distribution based on non Gauss.This is mainly composed of three steps.The first is the application of Freeman-Durden decomposition of the pixel to divide the scattering into three types: surface scattering, volume scattering and dihedral scattering, and then by using the shape parameter the scattering type can be divided into three types.After that, the eight neighborhood priori probabilities for each pixel are calculated to improve the classification distance and calculate the cluster centers.Finally, the iterative K-Wishart classifier is applied to PolSAR image for accurate classification and the color padding scheme.Different from Wishart distribution, the K-wishart distribution is not only suitable for uniform regional data description, but also very strong for the general uneven regional data description.The experiment results show that the proposed method has better classification performance than Freeman-Durden decomposition and complex Wishart distribution.%结合Freeman-Durden以散射模型为基础开发的分解算法和基于非高斯的K-Wishart分布,提出了一种无监督算法对全极化合成孔径雷达(fully polarimetric SAR,PolSAR)数据进行地物分类.该算法主要由3大步骤组成: 首先通过Freeman-Durden算法把PolSAR数据划分成3种散射: 表面散射、体散射和二面角散射,再使用形状参数χ将各种散射分为3类;然后通过每个像元的8个邻域计算先验概率,以改进分类距离和计算聚类中心;最后应用迭代K-Wishart分类器进行精确分类,并对每一类提出颜色填充方案.与复Wishart分布不同,K-wishart分布不但适合均匀区域数据描述,而且对不均匀区域数据的描述能力也很强.实验结果表明,该方法比Freeman-Durden分解和复Wishart分布组合具有更好的分类性能.