欢迎访问中国科学院大学学报,今天是

中国科学院大学学报 ›› 2012, Vol. 29 ›› Issue (1): 62-69.DOI: 10.7523/j.issn.2095-6134.2012.1.009

• 信息与电子科学 • 上一篇    下一篇

基于核函数原型和自适应遗传算法的 SVM模型选择方法

陈刚, 王宏琦, 孙显   

  1. 中国科学院电子学研究所空间信息处理与应用系统技术重点实验室, 北京 100190
  • 收稿日期:2010-11-26 修回日期:2011-01-28 发布日期:2012-01-15
  • 基金资助:

    国家自然科学基金(41001285)资助

Model selection for SVM classification based on kernel prototype and adaptive genetic algorithm

CHEN Gang, WANG Hong-Qi, SUN Xian   

  1. Key Laboratory of Technology in Geo-spatial Information Processing and Application System, Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China
  • Received:2010-11-26 Revised:2011-01-28 Published:2012-01-15

摘要:

针对现有SVM模型选择方法中人为指定核函数类型导致SVM模型性能难以达到最优的问题,提出了核函数原型的概念,并在此基础上提出一种基于核函数原型和自适应遗传算法的SVM模型选择方法. 该方法针对具体问题选择最优的核函数,有效地提高了SVM模型的性能;同时该方法通过动态调整遗传算法的控制参数,提高了SVM模型选择方法的稳定性. 在5个标准SVM数据集和遥感图像上的实验证明了该方法的有效性和稳定性.

关键词: 支持向量机, 自适应遗传算法, 模型选择, 场景分类, 遥感图像

Abstract:

Model selection plays an important role in support of vector machine classification. We propose a concept of kernel prototype which means general kernel type. Based on this, a new algorithm for model selection for supporting vector machine classification is proposed. The powerful adaptive genetic algorithm is used to optimize the parameters in kernel prototype, and it can find the optimal kernel type as well as the kernel parameters. Experiments show effectiveness and stability of the algorithms.

Key words: support vector machine (SVM), adaptive genetic algorithms (AGA), model selection, scene classification, remote sensing image

中图分类号: