基于L1范数B2DPCA的SAR目标特征提取与识别

SAR target feature extraction and recognition based on L1-norm B2DPCA

摘要:

主成分分析法(PCA)等基于L2范数最小均方准则的目标特征提取方法在合成孔径雷达(SAR)图像目标识别中得到广泛应用，L2范数易受SAR图像中野值的干扰，影响目标特征提取效果。介绍一种基于L1范数双向二维主成分分析法(B2DPCA-L1)的目标特征提取方法。L1范数对野值有较强的鲁棒性，通过在L1范数框架下实现B2DPCA，有效地改善了样本中野值对特征提取的影响，同时减少了特征矩阵维数，提高了目标识别率。实验表明，所提出方法的识别性能优于基于L2范数的特征提取方法。

Abstract:

The Minimum Mean-Square based methods like Principal Component Analysis(PCA) are widely used in Synthetic Aperture Radar(SAR) target recognition. However the L2-norm based criteria is prone to be affected by the outliers, which is not good for the target feature extraction in SAR imagery. To solve this problem，a L1-norm based bilateral two Dimension Principal Component Analysis(B2DPCA-L1) is proposed. The L1-norm version of B2DPCA is robust to outliers, and reduces the dimension of feature matrix and improves the target recognition rate as well. Experiments show that the proposed method has a higher target recognition rate in SAR imagery compared with the traditional L2-norm based feature extraction methods.