两向2DLDA与SVM相结合的SAR图像识别

SAR Images Target Recognition Based on Bidirectional 2DLDA and SVM

针对线性判决分析(LDA)用于图像特征提取时存在破坏二维空间结构、特征向量维数过大的缺点。二维线性判决分析(2DLDA)直接对图像矩阵进行运算,在一定程度上弥补了LDA的缺陷,但其实质是按行压缩图像矩阵进行特征提取,只消除了图像列的相关性,所提取的特征维数依然过大。为解决以上问题,本文采用两向2DLDA的方法,在行和列方向同时压缩图像矩阵进行特征提取。并结合支持向量机(SVM)进行分类识别,用MSTAR计划发布的实测合成孔径雷达(SAR)图像数据进行实验。结果表明,该方法在减少计算量的同时能达到较高的识别率。

Aiming at the method of Linear Discriminant Analysis has the disadvantages of damaging the two dimension spatial structure and feature vectors oversize.Image matrix was computed directly by the 2-dimensional LDA,these flaws are recover in some way.However,the essence of 2DPCA is to extract features of image matrix in each row,the relativity between columns was eliminated,the dimensions of features is still oversize.In order to solve the above problems,in this paper,the method of bidirectional 2DLDA is adopted to extract features by compressing image matrix along the columns and rows.Experimental results performing on SAR ground stationary military target data was published by MSTAR,and the Support Vector Machine is chosen as classifier.The result shows that the method presented in this paper can decrease calculated amount and raise recognition rate.