基于特征参数稀疏表示的SAR图像目标识别  

Target recognition in SAR images using sparse representation based on feature space

作　　者：王燕霞[1] 张弓[1] 

机构地区：[1]南京航空航天大学电子信息工程学院,南京210016

出　　处：《重庆邮电大学学报（自然科学版）》2012年第3期308-313,共6页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：航空基金(2011ZC52034);教育部留学回国人员科研启动基金;江苏高校优势学科建设工程资助项目~~

摘　　要：通过对合成孔径雷达(synthetic aperture radar,SAR)图像的统计特性分析,提出一种基于特征参数稀疏表示的SAR图像目标识别算法,有效地解决了图像域稀疏表示识别算法存在的高维问题。由低维高精度的广义二维主分量特征构成过完备字典,基于Fisher线性判别准则对该字典进行学习优化,使得类内更紧凑,类间更分开,同时降低了稀疏求解的复杂度。求解测试样本在优化字典下的稀疏表示系数,根据系数矢量的能量特征完成分类识别。MSTAR(moving and stationany target acquisition and recognition)实测SAR图像数据实验的结果表明,该方法稀疏求解复杂度低。Pointing at the high dimension problem in SAR images＇ target recognition algorithm using sparse representation in image domain,we propose a new algorithm based on feature space after analyzing SAR images＇ statistical characteristic.First,the generalized 2-dimensional principal component analysis（G2DPCA） feature with low-dimension and high-precision is extracted to form an over-complete dictionary.Then,2D-Fisher linear discriminate criterion is used to optimize the dictionary,which makes correlation of atoms in the same class more compact and difference of atoms between classes more apart.Besides,optimization process cuts down complexity in sparse solving.Sparse representation coefficient of test sample is computed based on the optimal dictionary.Classification and recognition is realized according to the energy feature of coefficient.Experiment results based on MSTAR SAR image data show that,algorithm raised in this essay lowers complexity in sparse solving,and increases recognition accuracy and speed effectively within simple preprocessing of SAR images.

关 键 词：合成孔径雷达(SAR)图像 广义二维主分量分析(G2DPCA) 目标识别 稀疏表示 移动和静止目标获取与识别(MSTAR) 

分 类 号：TN97[电子电信—信号与信息处理] TP751[电子电信—信息与通信工程]