结合非采样剪切波和MCCA的SAR目标识别方法

SAR Target Recognition via Combination of NSCT and MCCA

合成孔径雷达(SAR)图像处理是获取侦察信息的重要手段,当前目标识别能力不高已成为制约SAR有效获取侦察信息的关键问题。针对这一问题,提出结合基于非下采样contourlet变换(NSCT)和多重集典型相关分析(MCCA)的SAR图像特征提取方法并据此设计目标识别算法。首先,基于NSCT对SAR图像进行多层次分解,在不同尺度上获得SAR图像的表征结果;基于MCCA在各个分解尺度上对获取结果进行融合处理,形成对应的特征矢量;然后,以联合稀疏表示为多任务学习的基础工具,对不同尺度上的融合特征矢量进行分析;最后,根据不同尺度特征矢量的结果获取识别结果。实验采用MSTAR数据集为基础素材,对提出方法进行能力测试和结果评估,验证了该方法的有效性。

Synthetic aperture radar(SAR)image processing was an important means to obtain reconnaissance information.The current low target recognition ability has become a key problem restricting the effective acquisition of reconnaissance information by SAR.To solve this problem,a SAR image feature extraction method based on nonsubsampled contourlet transform(NSCT)and multiset canonical correlation analysis(MCCA)was proposed,and a target recognition algorithm was designed accordingly.Firstly,the SAR image is decomposed based on NSCT,and the characterization results of SAR image were obtained at different scales.On this basis,the obtained results were fused on each decomposition scale based on MCCA to form the corresponding feature vector.Then,using joint sparse representation as the basic tool of multi task learning,the fused feature vectors at different scales were analyzed.Finally,according to the reconstruction errors of different scale feature vectors,the target category of SAR image was determined.The experiments took the MSTAR dataset as the basic material,which designed a variety of conditions to test the ability of the proposed method and evaluate the results,whose results validated the effectiveness of the proposed method.