电子学报 ›› 2019, Vol. 47 ›› Issue (8): 1708-1716.DOI: 10.3969/j.issn.0372-2112.2019.08.014

• 学术论文 • 上一篇    下一篇

基于简约凸壳的一类模糊支持向量机

周国华1,2, 卢剑炜1, 顾晓清3, 殷新春2   

  1. 1. 常州工业职业技术学院信息工程系, 江苏常州 213164;
    2. 扬州大学信息工程学院, 江苏扬州 225127;
    3. 常州大学信息科学与工程学院, 江苏常州 213164
  • 收稿日期:2018-11-20 修回日期:2019-03-22 出版日期:2019-08-25
    • 通讯作者:
    • 周国华
    • 作者简介:
    • 卢剑炜 男,1982年出生,江苏靖江人,硕士,现常州工业职业技术学院信息工程系副教授,主要研究领域为智能信息处理.E-mail:ljw@czili.edu.cn;顾晓清 女,1981年出生,江苏常州人,博士,现常州大学信息科学与工程学院硕士生导师,研究方向为模式识别,模糊系统.;殷新春 男,1962年出生,江苏姜堰人,博士,教授,现扬州大学博士生导师.研究方向为信息安全,软件质量保障、高性能计算.
    • 基金资助:
    • 国家自然科学基金 (No.61472343,No.61806026); 江苏省自然科学基金 (No.BK20180956); 院创新团队项目 (No.YB201813101005)

One-Class Fuzzy Support Vector Machine Based on Reduced Convex Hull

ZHOU Guo-hua1,2, LU Jian-wei1, GU Xiao-qing3, YIN Xin-chun2   

  1. 1. Department of Information Engineering, Changzhou Institute of Industry Technology, Changzhou, Jiangsu 213164, China;
    2. College of Information Engineering, Yangzhou University, Yangzhou Jiangsu 225127, China;
    3. School of Information Science and Engineering, Changzhou University, Changzhou, Jiangsu 213164, China
  • Received:2018-11-20 Revised:2019-03-22 Online:2019-08-25 Published:2019-08-25
    • Corresponding author:
    • ZHOU Guo-hua
    • Supported by:
    • National Natural Science Foundation of China (No.61472343, No.61806026); Natural Science Foundation of Jiangsu Province (No.BK20180956); Innovation Team Project for Changzhou Vocational Institute of Light Industry (No.YB201813101005)

摘要: 为解决传统一类支持向量机对噪声数据敏感和不适用于大规模分类等问题,提出了用于大规模噪声环境的基于简约凸壳的一类模糊支持向量机(OC-FSVM-RCH).OC-FSVM-RCH根据简约凸壳的定义在核空间得到代表正常类数据几何特征的样本,然后基于改进的模糊支持向量域描述算法,使得正常类数据包含在最小超球内,异常数据与超球间隔最大化.OC-FSVM-RCH剔除正常类数据轮廓边缘处的噪声,同时对数据内部的噪声不敏感.实验结果表明了所提算法在性能和训练时间上取得了良好的效果.

关键词: 模糊支持向量机, 一类分类, 简约凸壳, 噪声数据

Abstract: The traditional one-class support vector machines are sensitive to noise data and not suitable for large-scale classification. In order to solve the problem, a novel one-class fuzzy support vector machine based on reduced convex hull called OC-FSVM-RCH is proposed for large-scale noise data classification. According to the reduced convex hull, OC-FSVM-RCH obtains the samples representing the geometric characteristics of normal class data in the kernel space. Then OC-FSVM-RCH improves the fuzzy support vector domain description algorithm, in which normal class data is enclosed in the smallest hypersphere,and the margin between abnormal class data and hypersphere is maximized. OC-FSVM-RCH can eliminate the noise at the edge of normal data contour and is insensitive to the noise inside the normal data. Experimental results show that the proposed algorithm achieves good results in terms of performance and training time.

Key words: one-class, fuzzy support vector machine, reduced convex hull, noise data

中图分类号: