kPCA特征提取算法的自动目标识别

Automatic target recognition based on kPCA feature extraction algorithm

分析了主成分分析(PCA)与核主成分分析(kPCA)的基本原理,比较了两者在处理数据方面的性能,得出了kPCA比PCA在处理非线性可分数据方面具有优势的结论.依据几何绕射理论(GTD),通过Matlab仿真方法得到HRRP(高分辨距离像)数据,并以这些数据作为训练和测试样本,结合SVM分类方法,分别测试比较了基于4种不同核函数的分类识别性能,得出基于高斯核函数主成分分析的自动目标识别系统性能明显好于其他3种核函数的结论.

This paper analyzed the basic principles of PCA （principal component analysis） and kPCA （kernel principal component analysis）, compared the performances of kPCA and PCA in data processing, and then drew a conclusion that kPCA has more advantages than PCA in dealing with nonlinear separable data. Based on geometrical theory of diffraction （ GTD）, the HRRP （ high resolution range profile） data were obtained by Matlab simulation method and used as the training and test sample. And then, combining with SVM classification method, performances of automatic target recognition system based on four different kernel functions were tested respectively. The ATR （ automatic target recognition） system based on principal component analysis of Gaussian kernel function was obviously better than the other three ones.