基于Randomfaces与稀疏表示的SAR目标识别

SAR Target Recognition Based on Randomfaces and Sparse Representation

为了准确地进行SAR图像目标识别,提出一种基于Randomfaces与稀疏表示的SAR目标识别方法,该方法首先利用Randomfaces进行训练样本的降维处理,然后利用降维后的训练样本构建稀疏线性模型,通过1范数最优化求解测试样本的稀疏系数解x,最后利用系数的稀疏性分布进行目标的分类识别。基于MSTAR数据进行了仿真验证,实验证明:基于Randomfaces与稀疏表示的SAR目标识别方法,在目标方位角未知的情况下识别率仍可达到98%以上,且Randomfaces的降维方式降低了在特征提取过程中对训练样本的要求。

In order to recognize SAR target accurately,a SAR ATR method based on randomfaces and sparse representation is proposed. Firstly,the training samples after dimensionality reduction using randomfaces are used to build a sparse linear model. Secondly,the sparse coefficient solution x of the test sample is solved by l1-minimization. Finally,the identification task is solved by utilizing the sparse distribution of the sparse coefficient. Experimental results with MSTAR dataset verify that the identification method based on sparse representation in a certain characteristic dimension can obtain good recognition performance. The recognition rate can reach more than 98 % without knowing the target azimuth,and the training samples are no longer critical in the feature extraction process because of the dimensionality reduction method of Randomfaces.