二维非参数最大散度差鉴别分析的SAR图像识别  

SAR Image Target Recognition Based on Two Dimensional Nonparametric Maximum Scatter Difference Discriminant Analysis

作　　者：姜晖[1] 刘振[1] 王鹏[1] 

机构地区：[1]解放军电子工程学院,合肥230037

出　　处：《火力与指挥控制》2014年第5期101-106,共6页Fire Control & Command Control

摘　　要：为增强线性鉴别分析(LDA)在图像识别中所提取特征的可鉴别性及避免小样本问题,提出了二维非参数最大散度差鉴别分析(2DNMSD)的图像特征提取方法。首先根据非参数特征分析的准则直接在二维图像矩阵上构造散布矩阵,然后基于最大散度差鉴别分析准则求取投影矢量。基于MSTAR计划录取的数据的仿真实验结果表明:即使方位角信息未知并且使用简单的最近邻分类器,该方法所提取特征在较低特征维数下的识别率也可以达到98%以上,表明了方法的有效性和正确性。To enhance the performance of linear discriminant analysis and avoid its＂small sample problem＂,an image feature extraction method based on two dimensional nonparametric maximum scatter difference（2DNMSD）discriminant analysis is proposed. In this method,the scatter matrices are constructed directly on the SAR image matrices by non-parametric feature analysis criterion. Then, projection vectors are computed using two dimensional maximum scatter difference discriminant analysis criterion. Experiments on MSTAR（Moving and Stationary Target Acquisition and Recognition） public database show that recognition rates of the new method can achieve more than 98 % with less feature dimensionality,which indicates the correctness and effectiveness of our method.

关 键 词：FISHER线性鉴别分析 最大散度差鉴别分析 非参数特征分析 合成孔径雷达 目标识别 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] TN951[自动化与计算机技术—控制科学与工程]