鲁棒核主元分析的数据重构

Data Reconstruction Based on Robust Kernel Principal Component Analysis

摘要: 针对野值点噪声对样本均值和协方差估计带来的不利影响，在线性鲁棒M位置估计方法的基础上，结合了核原理来估计协方差，提出了一种新型的鲁棒KPCA（核主元分析）算法．将所提出的算法应用于数据重构仿真实验，仿真测试结果表明当样本数据中存在野值点噪声时，由所提出的鲁棒KPCA算法实现样本数据重构时，要比KPCA具有更高的重构精度，抗野值点噪声性能更强．

Abstract: To avoid the adverse effects of outliers noise on the sample mean and covariance,a new robust kernel principal component analysis(R-KPCA) is presented with the covariance estimated by combining the linear robust M-estimation of location with the kernel function.The presented algorithm is employed in a data reconstruction,and the simulation results show that the de-noise performance of the presented robust KPCA is better than that of KPCA to process the outlier noise. Moreover,the presented algorithm for data reconstruction has a higher precision than KPCA.