中文摘要：

为降低合成孔径雷达（Synthetic aperture radar,SAR）图像目标识别中目标方位角的影响,并提高对SAR变形目标的识别率,本文提出了一种基于压缩感知和支持向量机决策级融合的目标识别算法。该算法首先基于稀疏表征理论将SAR目标识别问题描述为压缩感知的稀疏信号恢复问题,然后基于稀疏系数分别进行目标类别判别与方位角估计。对样本进行姿态校正后,利用支持向量机分别对经过姿态校正和未经姿态校正的样本进行目标分类。最后采用投票表决法对3种算法的分类结果进行决策级融合。实验结果表明,基于压缩感知结果进行目标方位角估计有效,且随着训练样本数的增加,提出的决策级融合算法提高了SAR变形目标的识别率。

英文摘要：

To reduce the influence of aspect angle to synthetic aperture radar （SAR） object recognition and improve recognition rate of SAR distorted object, the algorithm of compressed sensing （CS） and sup- port vector machine （SVM） decision fusion for SAR object recognition is proposed. SAR object recogni- tion is described as a sparse signal recovery problem in CS based on sparse representation theory, and an object classification result and an aspect angle are obtained through sparse coefficient separately. The classification results are obtained by SVM classifier using rectified and original samples after rectifying the pose of test sample. The final recognition result is obtained through fusion of the three above results based on majority vote. Experimental results demonstrate that, the algorithm of object aspect angle esti- mation based on compressed sensing result is effective, and the proposed decision fusion algorithm im- proves deformable object recognition rate significantly as the sample number increases.