结合NSCT和TPCA的SAR图像目标识别

Target Recognition of SAR Images Via Combination of NSCT and TPCA

提出了一种结合非下采样轮廓波变换(NSCT)和张量主成分分析(TPCA)的合成孔径雷达(SAR)图像目标识别方法。采用NSCT对SAR图像进行分解获得多尺度、多方向的子带图像,从而为目标提供更充分的描述信息。采用TPCA对各个子带图像进行特征提取,降低其中的冗余。基于各个子带图像的特征矩阵,通过线性加权的方法获得测试样本与训练样本之间的距离测度。根据K近邻(K-NN)的基本思想对测试样本进行分类决策。采用MSTAR数据集设置多种实验条件对提出方法进行测试,结果反映了该方法的有效性。

A synthetic aperture radar(SAR)target recognition method is proposed via combination of nonsubsampled Contourlet transform(NSCT)and tensor principal component analysis(TPCA).First,NSCT is used to decompose SAR images into multi-scale and multi-direction subband images,which provide more descriptions for the targets.Afterwards,TPCA is performed on the subband images to reduce the redundancy.Based on the feature matrices from the subband images,a distance measure is designed via linear weighting fusion.Finally,the test sample is classified based on the idea of K-nearest neighbor(K-NN)according to the designed distance measure.Several experiment conditions are designed based on the MSTAR dataset to test the proposed method.The results show its effectiveness.