高分辨率极化SAR影像KummerU分布非监督分类方法

针对高分辨率极化SAR数据特征分布不再符合同质区域假设, 进而导致基于统计分布的极化SAR影像非监督分类方法精度下降的问题, 将具有广泛适用性的KummerU分布嵌入粒子群寻优聚类算法, 提出了新的极化SAR影像非监督分类算法(PSO-KummerU方法):首先基于极化SAR统计特征对数据进行初分类, 然后采用极化SAR统计特征与粒子群优化算法进一步进行聚类中心求解, 分类准则部分采用KummerU距离改进代替传统的Wishart距离度量准则; 采用3种非监督分类方法(H/α-Wishart、PSO-Wishart、PSO-KummerU方法)进行分类对比实验.实验结果表明:基于KummerU分布的PSO-KummerU方法与采用Wishart距离的聚类方法相比, 目视效果明显改进, 整体分类精度提高14%以上.

The Unsupervised Classification Method with KummerU Distribution of High Resolution PolSAR Images

The feature distribution of high-resolution polarimetric SAR data no longer conforms to the hypothesis of homogeneous region, which leads to the decline of unsupervised classification accuracy of polarimetric SAR image based on statistical distribution. A novel unsupervised classification algorithm for PolSAR image is proposed by embedding the widely applicable KummerU distribution into the Particle Swarm Optimization clustering algorithm. Firstly, the PolSAR data was classified based on the polarimetric statistical characteristics. Then combining the PolSAR statistical characteristics with PSO algorithm, a further solution to the clustering centers was found with the PSO-KummerU algorithm. In the part of the classification criteria, KummerU distance was used to replace the traditional Wishart distance to improve the classification result. Finally, 3 kinds of unsupervised classification methods (H/α-Wishart, PSO-Wishart, PSO-KummerU) were used for the comparison experiments. The experimental results show that the visual effect of PSO-KummerU clustering method based on KummerU distribution is significantly improved compared with the Wishart distance clustering method, and the overall classification accuracy is improved by more than 14%.