基于小波字典稀疏表示的SAR图像目标识别

Target Recognition of SAR Images Based on Sparse Representation of Wavelet Dictionary

针对稀疏表示识别算法在图像域构造冗余字典时过分依赖预处理及原子维数较大的问题,提出基于小波字典的SAR图像稀疏表示识别算法。首先采用二维离散小波变换将原始图像变换到小波域,建立小波域SAR图像特征模型,得出小波域低频成分可充分表征目标类别信息的结论。然后取小波域低频成分进行2DPCA特征抽取构造小波字典,最后由改进OMP算法稀疏分解系数得到识别结果。SAR MSTAR数据的实验结果表明,在无预处理的情况下识别率高达99%,并且在含噪比10%的情况下识别率仍达96%。

This paper proposes a sparse representation classification（SRC）algorithm of SRA images based on wavelet dictionary.The proposed algorithm can solve the problems of high atom dimensionality and over-dependence on preprocessing when a redundant dictionary is being created by SRC algorithm in image domain.First,original image is transformed into wavelet domain by two-dimensional discrete wavelet trans-formation.And the SAR image feature model of wavelet domain is constructed and a conclusion is drawn that only low frequency component fully represents the target information.Then a wavelet dictionary is construc-ted through 2DPCA feature extraction by using the low-frequency component.Finally,the recognition result is obtained by sparse decomposition coefficients with the modified OMP algorithm.Experiment result based on MSTAR SAR image data shows that the recognition rate is as high as 9 9% without image preprocessing, and even in case of noise ratio of 10% the recognition rate is still 96%.