特征有效提取的自适应核特征子空间方法  

Adaptive Kernel Feature Subspace Method for Efficient Feature Extraction

作　　者：张朝阳[1] 田铮[1] 

机构地区：[1]西北工业大学理学院

出　　处：《模式识别与人工智能》2013年第4期392-401,共10页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金项目(No.10926197;60972150);中国留学基金委研究生项目(No.2011629111)资助

摘　　要：基于核的主成分分析(KPCA)方法能提取数据的非线性特征,但特征提取的效率却与训练样本集合的容量成反比.文中提出一种特征提取的自适应核特征子空间方法来快速有效地提取特征.该方法和KPCA方法在理论分析框架上是一致的,但通过自适应的选取核子空间的张成向量,能在提高特征提取效率的同时不影响特征提取的精度.针对模拟数据和MNIST数据的实验结果表明文中方法优于经典KPCA方法和参考方法.Kernel principal component analysis （KPCA） can extract nonlinear features of datasets. However, its efficiency is inversely proportional to the size of the training sample set. In this paper, an adaptive kernel feature subspace method is proposed to extract features efficiently. This method is methodologically consistent with KPCA, and it improves the efficiency by adaptively selecting the spanning vectors of the KPCA without losing accuracy. Experimental results on two-dimensional data and MNIST datasets show that the proposed method is better than the one associated with KPCA and reference methods.

关 键 词：核主成分分析(KPCA) 特征提取 核子空间 张成向量 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]