电子学报 ›› 2018, Vol. 46 ›› Issue (5): 1246-1252.DOI: 10.3969/j.issn.0372-2112.2018.05.034

• 科研通信 • 上一篇    下一篇

动态模糊密度的多分类器融合算法

李艳秋, 任福继, 胡敏   

  1. 1. 合肥工业大学计算机与信息学院, 安徽合肥 230009;
    2. 情感计算与先进智能安徽省重点实验室, 安徽合肥 230009
  • 收稿日期:2017-03-13 修回日期:2017-06-30 出版日期:2018-05-25
    • 作者简介:
    • 李艳秋 女,1988年生于安徽淮北.合肥工业大学计算机与信息学院在读博士生.研究方向为情感计算、多分类器集成.E-mail:liyanqiu2012@163.com;任福继 男,1959年生于四川南充.合肥工业大学计算机与信息学院教授,博士生导师.研究方向为情感计算、自然语言处理、人工智能等.Email:ren2fuji@gmail.com
    • 基金资助:
    • 国家自然科学基金 (No.61672202,No.61432004,No.61502141); 国家自然科学基金-深圳联合基金重点项目 (No.U1613217)

Dynamic Fuzzy Density for Multi-classifier Fusion Algorithm

LI Yan-qiu, REN Fu-ji, HU Min   

  1. 1. School of Computer and Information, Hefei University of Technology, Hefei, Anhui, 230009, China;
    2. Affective Computing and Advanced Intelligent Machines Anhui Key Laboratory, Hefei, Anhui 230009, China
  • Received:2017-03-13 Revised:2017-06-30 Online:2018-05-25 Published:2018-05-25
    • Supported by:
    • National Natural Science Foundation of China (No.61672202, No.61432004, No.61502141); Key Program of NSFC-Shenzhen Joint Fund (No.U1613217)

摘要: 在分析现有模糊密度计算方法的基础上,本文从分类器的隶属度分布和输出一致性两方面探索计算模糊密度的新方法,提出一种基于决策信任度和支持度的动态模糊密度赋值方法,旨在根据各分类器识别具体目标时输出的客观信息,实时地刻画分类器在融合系统中的可靠性.在表情识别上的实验结果表明,本文方法可以有效提高模糊积分融合的决策性能,降低单分类器输出不可靠决策信息的干扰,是一种有效的多分类器融合方法.

关键词: 多分类器融合, 模糊积分, 模糊密度, 表情识别

Abstract: Based on the analysis of the existing fuzzy density calculation methods,this paper explores a new method of calculating fuzzy density from the membership degree distribution and output consistency of the classifiers,and proposes a dynamic fuzzy density assignment method based on decision trust and support degree,which aims to describe the reliability of the classifier in the fusion system in real time according to the objective information output when each classifier identifies the specific target.The experimental results on facial expression recognition show that the proposed method can effectively improve the decision performance of fuzzy integral fusion and reduce the interference of unreliable decision information output by single classifier,which is an effective multi-classifier fusion method.

Key words: multi-classifier fusion, fuzzy integral, fuzzy density, facial expression recognition

中图分类号: