基于推广能力测度的多类SVDD模式识别方法.pdfVIP

3 Vol. 37 No. 3 009 3 ACTA ELECTRONICA SINICA Mar. 009 SVDD 朱孝开, 杨德贵 ( , 410073) : SVDD( Support Vector Domain Description) ( ) SVC ., , SVDD . , SVDD , , SVC . : ; SVDD; ; : TP3914 : A : 037 11 ( 009) MultiClass Support Vector Domain Descriptionfor Pattern Recognition Based on a Measure of Expansibility ZHU Xiaokai,YANG Degui ( Research Insti tute of Sp ace Electronics Inf ormation Technol ogy , College of Electric Science and Engineering , N UDT , Changsha , H unan 410073, China) Abstract: Classic SVDD classifiers, which use distance measures, have lower recognition rate than normal twoclass SVM classifiers. After analyzing the causes, a new measure is proposed, which can represent the more essential relationship between sam ples and categories.And then, a multilevel SVDD is proposed.The experiment on real onedimensional range profiles data show s that, the multi level SVDD reserves the lower complexity, higher expansibility and fewer requirements to sample size, ERs are re duced effectively, CR is increased even to the level of twoclass SVM classifiers. Key ords: pattern recognition; SVDD; multi level architecture; multi class classification 1 2 SVDD [ 1] 1999 , David M. J.Tax SVM 21 SVDD 6] ( Support Scholkopf SVDD , Vector Domain Description, SVDD) [ ] ,