基于压缩感知的SAR图像目标识别

摘要

An approach based on compressed sensing is presented for synthetic aperture radar images target recognition, which converts target recognition into approximate solution solving of sparse representation. This approach can be used to implement approximate sparse representation of samples by using sparsity of test sample based on whole training samples. By reviewing distribution characteristic of sparse coefficient mainly focusing over real sort of sam- ples, discriminability of sparse coefficient itself on target sort is studied, and finally, target recognition is accom- plished by designing sorting algorithm on basis of distribution characteristic of sparse coefficient. The experiment based on three kinds of target of MSTAR data proves that using this approach can obtain much higher recognition rate comparing with several kinds of available typical approaches and it is an effective approach for SAR images target recognition.%提出一种基于压缩感知的合成孔径雷达图像目标识别方法，将目标识别问题转化为稀疏表示的近似求解问题。该方法利用测试样本在全体训练样本基下的稀疏性，实现样本间的近似稀疏表示。通过考察稀疏系数主要集中于样本真实类别之上的分布特性，研究了稀疏系数本身对目标类别具有的可区分能力，最后基于稀疏系数的分布特性设计分类算法完成目标识别。基于MSTAR数据中三类目标的实验证明，与目前已有的几种典型方法相比，该方法可以取得更高的识别率，是一种有效的合成孔径雷达图像目标识别方法。