基于SVM与非参数LDA的雷达自动目标识别  

Radar automatic target recognition based on SVM and nonparametric LDA

作　　者：刘敬[1] 张军英[1] 赵峰[1] 

机构地区：[1]西安电子科技大学计算机学院,西安710071

出　　处：《控制与决策》2007年第11期1250-1254,共5页Control and Decision

基　　金：国家自然科学基金项目(60574039;60371044)

摘　　要：针对非参数线性判别分析(LDA)的类间散布矩阵,就如何有效描述类边界结构这一问题,提出一种SVM与k近邻(kNN)法相结合的非参数类间散布矩阵构造方法——SVM-kNN.该方法消除了非类边界样本对类边界结构信息的扭曲.将SVM-kNN非参数LDA方法用于外场实测高分辨距离像的特征提取,并将识别结果与加权kNN非参数LDA法和谱域原空间法比较,结果表明,SVM-kNN非参数LDA方法能显著提高识别效率.The between class scatter matrix of nonparametric linear discriminant analysis（LDA） faces the problem of how to efficiently specify the boundary structure of original data. A method of constructing nonparametric between class scatter matrix by combining SVM and k nearest neighbors （kNN） method, SVM-kNN method, is presented. The presented method avoids the distortion of boundary structure information caused by non-boundary samples. The presented SVM-kNN nonparametric LDA method is applied to the feature extraction of outfield real high resolution range profiles. Comparing with weighted kNN nonparametric LDA method and original spectrum domain method, the recognition results show that the presented SVM-kNN nonparametric LDA method can improve recognition efficiency significantly.

关 键 词：雷达自动目标识别 非参数线性判别分析 支撑向量机 特征提取 高分辨距离像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN957.52[自动化与计算机技术—计算机科学与技术]