以非负矩阵分解提取局部特征的SAR目标稀疏表示分类

Sparse Representation Classification of SAR Targets with Local Features Extracted by Non-negative Matrix Factorization

合成孔径雷达(SAR)目标分类是自动目标识别系统的核心功能之一,对于战场监视等应用具有重要意义。利用SAR图像局部散射明显的特点,提出了通过训练样本的非负矩阵分解获得低维数局部特征编码,并以该编码作为字典进行稀疏表示分类的方法。采用Gotcha项目民用车辆目标的实测数据进行了验证,结果显示在不同信噪比条件下该方法的分类正确率均优于广泛采用的由降采样、随机投影、主成分分析提取低维数特征的稀疏表示分类方法,表明了该方法的性能优势。另外,还通过实验对比分析了非负约束的稀疏表示与标准稀疏表示在分类性能上的差别,结果显示非负约束的稀疏表示导致分类正确率下降,故针对分类问题不宜在稀疏表示时进行非负约束。

Synthetic aperture radar( SAR) target classification is one of the core functions in automatic tar-gets recognition( ATR) system. It is essential in battle field surveillance,too. According to the characteris-tics that SAR images have prominent local scattering,it is proposed to perform non-negative matrix factori-zation( NMF) on the training samples to get low dimensional local encoding matrix,and subsequently per-form sparse representation classification( SRC) based on this encoding matrix. Processing results on real data of civilian vehicle targets in Gotcha project demonstrate that the proposed method outperforms other di-mension reduction methods such as down-sampling,random projection and principle components analysis, which are adopted with SRC. In this way,superiority of the method is revealed. Besides,the performances of SRC with and without non-negativity constraints are compared and analyzed by experiments. The experiment result reveals that SRC with non-negativity constraints leads to degradation of classification performance. In this way,it is unadvisable to include non-negativity constraint with regard to classification problem.