基于目标分解及纹理信息的全极化SAR影像分类  

Full Polarimetric SAR Image Classification Based on Target Decomposition and Texture Information

作　　者：刘雨思 余洁[1] 张晶[1] 

机构地区：[1]首都师范大学资源环境与旅游学院,北京100048

出　　处：《地理空间信息》2018年第4期11-13,17,共4页Geospatial Information

基　　金：国家自然科学基金资助项目(41671417)

摘　　要：全极化SAR影像含有丰富的特征信息,针对单一特征用于分类难以达到满意精度的问题,提出一种基于不同目标分解方法及纹理信息的SVM(Support Vector Machine,SVM)全极化SAR影像监督分类方法。结果表明,Cloude分解和Yamaguchi分解在极化特征信息提取时各有优势,且都优于Freeman分解效果;Cloude分解和Yamaguchi分解结合作为极化特征信息时,分类总体精度相对较高;纹理信息与极化特征信息在表现地物特性方面具有互补性,结合纹理信息后,分类总体精度提高了4.92%,为90.86%,Kappa系数为0.8754。Full polarimetric SAR image contains a wealth of information, but using single characteristics to classify is difficult to achieve satisfactory classification accuracy. Aiming at this problem, we proposed a SVM full polarimetric SAR image supervised classification method based on different target decomposition methods and texture information. The result shows that Cloude decomposition and Yamaguchi decomposition have their own advantages in the extraction of the polarization feature information, and both of them are superior to Freeman decomposition;Cloude decomposition and Yamaguchi decomposition have their own advantages in the extraction of the polarization feature information, and both of them are superior to Freeman decomposition. Combined with Cloude decomposition and Yamaguchi decomposition as the polarization feature information, the overall accuracy of the classification is relatively high. Texture information and polarization feature information are complementary to each other, with texture information, overall accuracy is improved by 4.92%, gives it of 90.86%, Kappa coefficient of 0.875 4.

关 键 词：全极化SAR 目标分解 纹理信息 影像分类 

分 类 号：P237[天文地球—摄影测量与遥感]