稀疏表示框架下的SAR目标识别

SAR Target Recognition under the Framework of Sparse Representation

稀疏表示选择最佳线性表示重构信号,可避免合成孔径雷达(SAR)目标识别中的方位角估计难题,同时减轻强相干噪声影响。稀疏字典选择是稀疏表示中的关键问题之一,该文提出分别使用级联方式和并联方式构造稀疏字典实现SAR目标识别。首先对训练样本进行对数归一化处理,使用主成分分析(PCA)特征提取和降维;然后对处理后的数据分别组成级联字典和并联字典,采用截断牛顿内点法(TNIPM)获得目标的稀疏表示;最后,在两种字典的稀疏表示框架下设计分类器对SAR目标识别。通过对比实验,验证了该文的字典构建方式在稀疏表示框架下对SAR目标识别的有效性。

Sparse representation uses only a few coefficients to linearly reconstruct the signal, which can avoid estimating the azimuth in synthetic aperture radar （SAR） target recognition as well as mitigate the impacts of strong coherent noises. The construction of the dictionary is a crucial issue in SAR target recognition under the framework of sparse representation. To improve the performance of SAR target recognition, the concatenated way and the parallel way are proposed to construct the dictionary for sparse representation. Firstly, the training samples are processed by the logarithmic transformation and then they are normalized. Moreover, principal components analysis （PCA） is employed to extract feature ~d reduce dimension. Secondly, the concatenated dictionary and the parallel dictionary are constructed, respectively. At last, the sparse representation of the SAR image is obtained by the truncated Newton interior-point method（TNIPM） with two different dictionaries, respectively. The testing sample belongs to the class with the minimum reconstruction error under the framework of the parallel dictionary while the class with the maximum coefficients sum under the framework of the concatenated dictionary. The experimental results demonstrate the effectiveness of our proposed algorithms.