引用本文:雷英杰,余晓东,王睿,王毅.根据混合选择策略的直觉模糊核匹配追踪集成算法[J].控制理论与应用,2016,33(3):336~343.[点击复制]
LEI Ying-jie,YU Xiao-dong,Wang Rui,Wang Yi.Intuitionistic fuzzy kernel-matching pursuit ensemble algorithm based on hybrid selection strategy[J].Control Theory and Technology,2016,33(3):336~343.[点击复制]
根据混合选择策略的直觉模糊核匹配追踪集成算法
Intuitionistic fuzzy kernel-matching pursuit ensemble algorithm based on hybrid selection strategy
摘要点击 2601  全文点击 2124  投稿时间：2015-04-02  修订日期：2015-09-24
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2016.50266
  2016,33(3):336-343
中文关键词  直觉模糊匹配追踪  选择性集成  混合选择策略  差异性度量  泛化性能
英文关键词  intuitionistic fuzzy kernel matching pursuit  selective ensemble  hybrid selection strategy  diversity measure  generalization performance
基金项目  国家自然科学基金项目(61272011, 61309022)资助.
作者单位E-mail
雷英杰 空军工程大学防空反导学院  
余晓东* 空军工程大学防空反导学院 agosoa@163.com 
王睿 空军工程大学防空反导学院  
王毅 空军工程大学防空反导学院  
中文摘要
      为了从分类器集成系统中选择出一组差异性大的子分类器, 从而提高集成系统的泛化能力, 提出了一种基 于混合选择策略的直觉模糊核匹配追踪算法. 基本思想是通过扰动训练集和特征空间生成一组子分类器; 然后采 用k均值聚类算法将对所得子分类器进行修剪, 删去其中的冗余分类器; 最后根据实际识别目标动态选择出较高识 别率的分类器组合, 使选择性集成规模能够随识别目标的复杂程度而自适应地变化, 并基于预期识别精度实现循环 集成. 实验结果表明, 与其他常用的分类器选择方法相比, 本文方法灵活高效, 具有更好的识别效果和泛化能力.
英文摘要
      In order to improve the generalization ability of a classifier ensemble, we propose an intuitionistic fuzzy kernel-matching pursuit algorithm based on the hybrid selection strategy for target recognition to select a subset of optimal individuals from the given classifier ensemble. The basic idea of this algorithm is to produce a preliminary subset of classifiers by disturbing the training set and the feature space, and then trim this subset by eliminating the redundant classifiers based on k-means clustering algorithm and dynamically singling out classifiers with high differentiability from practical object recognition, making the size of the subset adaptively change according to the complexity of the objects and the expected accuracy of recognition be determined recursively. Experimental results show that the performance of the proposed algorithm is more flexible, efficient and accurate, with higher generalization, in comparison to other classifier selection methods.