一种新的基于Wishart MRF的全极化SAR图像分类方法

无监督的Wishart分类算法在多次迭代后,容易出现错分现象,即多个类别属于同一类散射机制,或者多种散射都拥有相同的类别标签。针对此问题,提出了一种新的基于Wishart MRF的无监督全极化SAR图像分类方法。新方法改进了散射机制保持的方式,即并不是完全限制像素点的散射机制,而是根据像素点的散射机制在迭代过程中给定一个有限的范围。同时,使用一种自适应区域的MRF方法来提取像素点的先验信息。该方法不仅考虑了全极化SAR数据的散射性质,而且结合了统计特性和邻域信息,并在一定程度上保持了散射性质。实验结果证明,与传统的Wishart和基于散射机制保持的Wishart算法相比,该方法在JPL/NASA的AIRSAR数据上取得了更好的分类结果。

New Wishart MRF Method for Fully PolSAR Image Classification

The unsupervised Wishart classifier has usually given some misclassified pixels,i.e.,several classes present the same polarimetric scattering mechanism,or one class has several different polarimetric scattering mechanisms.Aiming at it,this paper proposed a new Wishart MRF method for fully polarimetric synthetic aperture radar (PolSAR) image classification.Instead of preserving the fixed scattering characteristic,the new method sets a constrained scope for the label shifting from one to another.In addition,prior information is extracted by MRF method with an adaptive neighborhood.The physical scattering characteristic is considered,as well as the statistics information and the spatial information,and the physical scattering characteristics are preserved in a certain degree.Compared with traditional Wishart classifier and modified Wishart classifier preserving polarimetric scattering characteristics,the proposed method has better classification results on JPL/NASA AIRSAR data of San Francisco.