计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (18): 68-73.

• 大数据与云计算 • 上一篇    下一篇

多核支持向量数据描述分类方法研究

卢  明,刘黎辉,吴亮红   

  1. 湖南科技大学 信息与电气工程学院,湖南 湘潭 411201
  • 出版日期:2016-09-15 发布日期:2016-09-14

Research on multi-kernel support vector data description method of classification

LU Ming, LIU Lihui, WU Lianghong   

  1. School of Information and Electrical Engineering, Hunan University of Science and Technology, Xiangtan, Hunan 411201, China
  • Online:2016-09-15 Published:2016-09-14

摘要: 核函数、惩罚因子、核参数是影响支持向量数据描述(SVDD)分类方法分类效果的重要因素。研究了多核支持向量数据描述(MKSVDD)分类方法,给出了多核支持向量数据描述分类方法的实现步骤,基于banana数据集分析了惩罚因子和核参数对分类效果的影响,重点讨论了多核函数的权值对支持向量数据描述边界分布的影响。仿真实验结果表明,与单核支持向量数据描述分类方法相比较,多核支持向量数据描述分类方法的分类效果更佳,为实际应用时参数的选择提供了参考。

关键词: 模式识别, 支持向量数据描述, 多核方法, 最优超球半径, 参数选择

Abstract: Kernel function, penalty factor, kernel function parameters are important factors influencing the classification effects of the Support Vector Data Description method(SVDD). This paper studies the Multiple Kernel Support Vector Data Description(MKSVDD) classification method, furthermore, gives the classification implementation steps of multi-
kernel support vector data description method. It analyzes the impact of the penalty factor and kernel function parameters on classification results, mainly discusses the influence of the weight of multi-kernel function on boundary distribution of SVDD based on banana data set. The simulation results reveal that, compared with the single kernel support vector data description method of classification, multi-kernel support vector data description method of classification has better performance, provides the reference of parameters selection in the practical application.

Key words: pattern recognition, support vector data description, multi-kernel method, optimal hyper-sphere radius, parameter selection