基于多层聚类的多分类SVM快速学习方法  

Multi-classification SVM multi-clustering fast learning method

作　　者：张春艳[1] 倪世宏[1] 张鹏[1] 查翔[1] ZHANG Chun-yan;NI Shi-hong;ZHANG Peng;ZHA Xiang(College of Aeronautics and Astronautics Engineering, Air Force Engineering University, Xi'an 710038,China)

机构地区：[1]空军工程大学航空航天工程学院,陕西西安710038

出　　处：《计算机工程与设计》2017年第2期522-527,共6页Computer Engineering and Design

摘　　要：针对SVM大规模多类样本学习效率较低的问题,提出一种基于多层聚类的多分类SVM快速学习方法。采用自下而上的方式构建二叉树层次结构,根据所得层次结构,对每个节点对应的样本进行学习。学习时对训练样本进行首次聚类得到若干类簇,对其中只有一类样本的类簇提取中心点;对有两类样本的类簇,根据其混合度,对其正负类样本设定不同的聚类数,进行二次聚类,提取所得类簇中心点。整合上述步骤中提取的中心点作为约简后的样本,学习并得到子分类器。仿真结果表明,基于多层聚类的多分类SVM快速学习方法能够在保证较高分类准确率的前提下,大幅约简训练样本,有效提高学习效率。A multi-classification SVM fast learning method based on multi-clustering was presented to solve its low learning effi-ciency problem when using SVM processing large scale multi-class samples.Bottom-up method was used to set up binary tree hi-erarchy structure,according to achieved hierarchy structure,sub-classifier learnt from corresponding samples of each node.Dur-ing the learning process?several class clusters were generated after the first clustering of the training samples.Central points were extracted from those class clusters which just had one type of samples.For those which had two types of samples?cluster numbers of their positive and negative samples were set respectively according to their mixture degree,secondary clustering was undertaken afterwards.Central points were extracted from achieved sub-class clusters.By learning from the reduced samples formed by the integration of extracted central points above,sub-classifiers were obtained Simulation results show that,this fast learning method can guarantee higher classification accuracy,greatly reduce sample numbers and effectively improve learning efficiency.

关 键 词：支持向量机 大规模训练集 多分类 多层聚类 二叉树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]