联合对称不确定性ReliefF算法的PolSAR影像分类  

PolSAR Image Classification with Combined Symmetric Uncertainty and ReliefF Algorithm

作　　者：张继超[1,2] 邹勇 宋伟东[1,2] 张永红[3] 李建飞 ZHANG Jichao;ZOU Yong;SONG Weidong;ZHANG Yonghong;LI Jianfei(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;Geospatial Information Service Collaborative Innovation Research Institute,Liaoning Technical University,Fuxin,Liaoning 123000,China;Chinese Academy of Surveying and Mapping,Beijing 100036,China;Liaoning Basic Institute of Surveying and Mapping,Liaoning Natural Resources Affairs Service Center,Jinzhou,Liaoning 121000,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]辽宁工程技术大学地理空间信息服务协同创新研究院,辽宁阜新123000 [3]中国测绘科学研究院,北京100036 [4]辽宁省自然资源事务服务中心辽宁省基础测绘院,辽宁锦州121000

出　　处：《遥感信息》2021年第4期20-27,共8页Remote Sensing Information

基　　金：国家自然科学基金项目(42071343、41271430、40971225);国家863计划资助项目(2011AA120404)。

摘　　要：针对在PolSAR影像分类中极易产生分类精度随着特征数增加不会持续增加,甚至还会降低的问题,提出一种基于对称不确定性ReliefF算法的分类方法。首先,在传统过滤式的ReliefF算法基础上引入对称不确定性评估函数,淘汰对分类贡献小的特征及属性;然后,利用封装式CART算法对剩余特征作进一步挑选,并根据得到的特征子集进行分类。将其与Wishart监督分类、未进行特征选择的分类和仅利用ReliefF算法进行特征属性选择的分类方法进行比较,以GF-3和Radarsat-2影像为例进行实验。结果表明,该方法各项指标均优于其他对比实验,并且大幅度节约时间成本。In order to solve the problem that the classification accuracy will not continue to increase or even decrease with the increase of feature number in PolSAR image classification,a classification method based on symmetric uncertainty evaluation function ReliefF algorithm is proposed.Firstly,based on the traditional filtered ReliefF algorithm,the symmetric uncertainty evaluation function is introduced to eliminate the features and attributes that contribute little to the classification.Then,the encapsulated CART algorithm is used to further select the remaining features and classify them according to the subset of features.It is compared with Wishart supervised classification,the classification method with no feature selection and the classification method only using ReliefF algorithm for feature selection.Taking the GF-3 and Radarsat-2 images as examples,the results show that each index of the proposed method is superior to that of comparative experiments,and the time cost is greatly saved.

关 键 词：极化分解 对称不确定性 RELIEFF算法 特征选择 CART分类 

分 类 号：TP722.6[自动化与计算机技术—检测技术与自动化装置]