基于SVM和能量最小化的PolSAR图像分类方法

极化合成孔径雷达(PolSAR)采用全极化的工作方式可以获取地物的多种特征,利用这些特征对地物进行分类是PolSAR图像的重要应用方向。不同的特征和分类器对分类精度有着较大的影响。提出了一种基于支持向量机(SVM)和能量最小化(EM)的极化SAR图像地物分类方法。该方法选择基于6种散射模型的分解方法(6SD)所得的6部分散射能量、总散射能量span和3个极化相干矩阵旋转域角参数作为SVM的输入,得到图像分类结果,并使用基于图割的能量最小化算法α-expansion对分类结果进行优化。最后使用AIRSAR系统获得的Flevoland地区的数据进行实验,结果表明所提算法可以提高总体分类精度,总体分类精度为95.6%,高于其他方法的92.3%。所提算法可以较大幅度地提高散射机理明显的区域,如建筑、森林、水域、草地等区域的分类精度。另外,结合EM优化结果可以提高所有种类的分类精度,其中在苜蓿、小麦1、小麦2、裸地、草地、油菜籽等区域的分类精度可提高1%以上。

Method of PolSAR image land cover classification based on SVM and energy minimization

Polarimetric synthetic aperture radar(PolSAR)can obtain a variety of features of ground objects by fully polarization working mode.It is an important application direction for PolSAR images to classify land cover by using these features.Different features and classifiers have great influence on classification accuracy.In this paper,a land cover classification method for PolSAR images based on support vector machine(SVM)and energy minimization(EM)is proposed.This method choose the six scattering models power obtained by model-based six-component decomposition(6 SD),the total scattering power and the angle parameters of polarimetric coherency matrix in rotation domain as the input of SVM.Using the EM algorithm based on graph cut,alpha-expansion,to optimize the SVM classification result,the method can improve the accuracy of classification.Finally,the data of Flevoland region obtained by AIRSAR system were used for experiments,and the results showed that the proposed algorithm could improve the overall classification accuracy,95.6%,higher than that of another method(92.3%).The proposed algorithm can greatly improve the classification accuracy of areas with obvious scattering mechanism,such as buildings,forests,waters and grasslands.In addition,the EM optimization results can improve the classification accuracy of all classes,among which the classification accuracy of Wheat 1,Wheat 2,Baresoil,Grassland and Rapeseed areas can be improved by more than 1%.