基于模糊积分和粒子群算法的多分类器融合.pdfVIP

您的论文得到两院院士关注 软件时空 文章编号:1008-0570(2010)03-3-0205-03 基于模糊积分和粒子群算法的多分类器融合 Multiple Classifiers Combination Based on Fuzzy Integral and Particle Swarm Optimization (长沙理工大学) 陈 曦 赵志伟 CHEN Xi ZHAO Zhi-wei 摘要 使用模糊积分的方法将多个分类器进行融合可以提高分类精度 但是如何得到最优的模糊测度是一个尚未解决的问 : , 题 本文根据模糊测度 积分的理想特性 用模糊测度代替各个分类器的权值 利用粒子群算法全局搜索的优势 将模 。 Sugeno , , , 糊测度对应于粒子 并随速度和位置并不断调整 从而得到全局最优的模糊测度 通过仿真实例验证了新的多分类器融合模 , , 。 型具有较低的分类错误率 并能有效地提高分类精度。 , 关键词 模糊积分 模糊测度 粒子群算法 全局最优 多分类器融合 : ; ; ; ; 中图分类号: TP391 文献标识码: A Abstract: Combination of many different classifiers can improve classification accuracy. Fuzzy integral is one of the ways to combine classifiers, but how to get the best fuzzy integral measure is a problem which remains unsolved. Considering the perfect characteristic of the Sugeno integral, this paper takes the fuzzy integral measure replacing the weight of the single classifiers. Based the advantage 技 on global search of the PSO, the fuzzy integral measure is adjusted along with the increase of the velocity and location of the parti- cle, the best global fuzzy integral measure is found finally. The experimental results show that this new model efficiently improve the 术 classifier accuracy. Key words: fuzzy integral; fuzzy measure; PSO; global optimi