二维压缩感知多投影矩阵特征融合的SAR目标识别方法

针对合成孔径雷达（synthetic aperture radar，SAR）目标识别问题，提出联合多层次二维压缩感知投影特征的方法。采用二维压缩感知投影作为基础特征提取算法，具有不依赖训练样本、效率高等显著优势。构建多个二维压缩感知投影矩阵提取原始SAR图像的多层次特征。不同投影矩阵获得的特征具有差异性，从不同方面描述SAR图像的灰度分布特性；同时，这些特征源自相同的输入图像，因此也具有一定的内在关联性。采用联合稀疏表示对提取的多个特征矢量进行表征分析，在内在关联性约束下考察不同特征的独立鉴别能力，从而提升每一类特征的稀疏表示精度。最终，根据求解的稀疏表示系数，分别在各个训练类别上对测试样本的多类特征进行重构，获得重构误差。根据最小误差的准则，判定测试样本所属目标类别。通过综合运用多层次二维压缩感知特征提取和联合稀疏表示分类，提高SAR目标识别的整体性能。利用MSTAR数据集中的多类目标SAR图像对方法进行测试验证，结果反映其在标准操作条件（standard operating condition，SOC）和扩展操作条件（extended operating condition，EOC）均可保持可靠的识别性能。

SAR target recognition using feature fusion by 2D compressive sensing with multiple random projection matrices

A synthetic aperture radar (SAR) target recognition was proposed using multi-layer projection feature based on 2D compressive sensing. 2D compressive sensing projection was employed as the basic feature extraction algorithm, which had the advantages of low dependency on the training samples, high efficiency, etc. Several projection matrices of 2D compressive sensing were constructed to extracted the multi-layer feature from original SAR images. The feature from different projection matrices had divergency, which reflected the gray distribution characteristics of SAR image from different aspect. Meanwhile, these feature came from the same input image, so they shared some inner correlation. Hence, the joint sparse representation was employed to classify the multi-layer feature, which could exploit their inner correlation to enhance the precision of each sparse representation problem. Finally, based on the solved sparse coefficients, the feature of the test sample was reconstructed on different training classes to obtain the reconstruction error. Based on the principle of the minimum reconstruction error, the target label of the test sample could be decided. The proposed method combined characteristics extraction of the multi-layer 2D compressive sensing and joint sparse representation classificaton to enhance the overall performance of SAR target recognition. The multi-class SAR images in the MSTAR dataset were used to test and validate the proposed method. The results confirm its reliable recognition performance under the standard operating condition(SOC) and extended operating conditions(EOC).