摘要
提出了对核主成分分析(KPCA)在模式分类中的特征提取的改进方法。KPCA对于模式分类数据,并不是投影后的主成分就是最利于分类的成分,因此把数据降维到一个利于分类的空间,同时剔除不利于分类的成分,尽可能保留类别信息,对于各个成分贡献率以及映射空间进行度量,依据各成分对于模式分类的贡献选择最优成分,且根据Fisher准则选择利于分类的空间,即确定类别信息量较大的成分以及KPCA的核参数。
For feature extraction in pattern classification using kernel principal component analysis (KPCA), a improved approach is proposed. To the pattern classification data, the principal components maybe be not the most beneficial for pattern classification. Reducing data dimensionality and eliminating bad components, it retains the categories information as much as possible. According to Fisher criteria and the contribution rate of the various components, the right space and components are selected, that is, to determine the major components according to the classification information and the kernel parameters of KPCA.
出处
《计算机工程与设计》
CSCD
北大核心
2010年第18期4085-4087,4092,共4页
Computer Engineering and Design
基金
重庆市科技攻关基金项目(CSTC
2009AB2049)
关键词
核主成分分析
分类
特征提取
FISHER
核
kernel principal component analysis
classification
feature extraction
Fisher
kernel