一种基于AFSA的SVM分类方法(英文)  

A classification method of SVM based on AFSA

作　　者：王卫星[1] 刘娟[1] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2011年第1期91-95,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：The National Natural Sciences Foundation of China(60873186)

摘　　要：应用一种全局搜索方法即人工鱼群算法(artificial fish swarm algorithm,AFSA)来优化支持向量基(support vector machines,SVM)的参数,并应用于图像分类。基于分类,初始化惩罚系数C和核函数参数δ2的范围;利用AF-SA来优化SVM的参数,并得到合适的值;最后,把参数优化后的SVM应用于分类。实验结果表明,与C-SVC和交叉验证法相比,其分类结果优于其它两种方法,因此AFSA-SVM方法有更好的准确性和鲁棒性。In this paper, artificial fish swarm algorithm （AFSA } that is a global search method to optimize the parameters of support vector machines （ SVM ） is applied and modified for image classification. In the classification, firstly, the range of parameters of punishment C and kernel function 62 are initialized ; secondly, AFSA is applied to optimize the parameters to gaiu suitable values; finally, SVM is used for classification, in which the parameters are optimized. By comparing with C-SVC and cross-validate methods, the result excelled another two methods, so the studied algorithm of AFSA-SVM is more accuracy and robust.

关 键 词：人工鱼群算法(AFSA) 支持向量基(SVM) C—SVC 交叉验证法 

分 类 号：TP301.5[自动化与计算机技术—计算机系统结构]