稀疏表示系数下局部最优重构的SAR图像目标识别算法

SAR Images Recognition Based on Sparse Coefficients Optimal Local Reconstruction

针对现有基于稀疏表示分类的合成孔径雷达(SAR)图像目标识别方法存在的不足,提出了稀疏表示系数下局部最优重构的SAR图像目标识别算法。该算法基于全局字典上求解得到的稀疏表示系数,选取若干具有最大系数的原子并分别在各个类别上对测试样本进行重构,获得相应的重构误差。同时,考虑到SAR图像方位角的敏感性,在各类别上选取与测试样本具有相近的方位角的原子,并对重构误差进行修正,从而获得最终的决策变量对目标类别进行判定。实验结果表明,该方法在标准操作条件和扩展操作条件下均可以取得更好的性能。

To solve the defaults in the present synthetic aperture radar(SAR)target recognition methods using sparse representation-based classification(SRC),a new method by searching optimal local reconstruction under global sparse representation was proposed.Based on the sparse coefficients over the global dictionary,only a few large coefficients were selected to reconstruct the test sample by different training classes.In addition,considering the azimuthal sensitivity of SAR images,the selected atoms from each class should have approaching azimuths with the test one.Therefore,when the atoms in each class were arranged according to the azimuths,the selected ones should gather in a small interval.Accordingly,the reconstruction errors were improved to obtain the final decision values to determine the target label.The experimental results showed that the proposed method could achieve superior performance under both the standard operating condition(SOC)and extended operating conditions(EOCs).