基于最小均值约束闭环树覆盖模型的一类分类器构造  

Research on One-class Classifier Algorithm Based on Closed-loop Tree Constrained by Minimum Mean Value Covering Model

作　　者：胡正平[1] 冯凯[1] 白洋[1] 

机构地区：[1]燕山大学信息科学与工程学院,秦皇岛066004

出　　处：《信号处理》2012年第10期1439-1445,共7页Journal of Signal Processing

基　　金：国家自然科学基金(61071199);河北省自然科学基金(F2010001297);第二批中国博士后基金特别资助(200902356)

摘　　要：传统的覆盖型一类分类器难以同时对数据主体和局部边缘结构进行准确描述。为此本文构造一种新的空间覆盖模型———闭环树结构。对数据主体流形,采用最小均衡树模型进行拟合,对局部边缘的多分枝结构以闭合环链模型链接包裹,形成平滑,明确的覆盖边缘。并通过最小均值约束原则实现两种模型的匹配对应,整合为一个整体。使数据流形的主干与边缘细节同时得到有效描述,多个数据库的实验也证明了该覆盖模型的合理性。Traditional One-class Classifier based on covering model is difficult to describe the main structure and specific edge of data manifold accurately at the same time.So this paper proposes a new space covering model-closed-loop tree tructure.It uses minimum balanced tree model to fit the main manifold of data,and uses closed chain model to link or enfold the multi-branch structure of local edge,forming a smooth,clear the edge of covering model.The Two models match each other by the constraint principle of minimizing mean,and integrated as a whole.It can describe the main structure and specific edge of data manifold effectively at the same time.The results of experiments based on several databases prove that the proposed method is good.

关 键 词：一类分类器 最小生成树 闭环树 覆盖模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]