基于混淆矩阵和集成学习的分类方法研究

摘要

For the multi-classification problem, a classification method based on confusion matrix and ensemble learning is proposed in this paper. A hierarchical structure is generated from the similarities between patterns. The classification method chooses support vector machine (SVM) as the basic binary classifier. The AdaBoost algorithm applies weighted voting on SVM whose classification accuracy is not ideal. Taking object recognition in the substation environmental monitoring as an example (related to people, animals, ordinary flames (red and yellow flames), white flames, incandescent lamps), object classification is achieved. The experiments show that the proposed method can effectively improve the classification accuracy.%针对多分类问题,本文提出一种基于混淆矩阵和集成学习的分类方法.从模式间的相似性关系入手,基于混淆矩阵产生层次化分类器结构；以支持向量机(SVM)作为基本的两类分类器,对于分类精度不理想的SVM,通过AdaBoost算法对SVM分类器进行加权投票.以变电站环境监控中的目标识别为例(涉及到人、动物、普通火焰(红黄颜色火焰)、白色火焰、白炽灯),实现了变电站环境监控中的目标分类.实验表明,所提出的方法有效提高了分类精度.