基于模式间混淆关系自适应构建层次化分类器

为了解决层次化分类器的设计难点——子分类器的层属关系及其内部组成的确定,本文首先定义了模式间混淆关系,用于描述不同模式在判决域中的相互作用;进而提出了基于混淆矩阵度量模式间混淆关系的方法。设计并实现了多模式混淆关系分析机MPCRAM,将有师指派和无师自组两种常用的模式重组方法有机结合,遵循Fisher准则,自适应地产生层次化分类器结构。大量综合测试证实了该方法有效、实用,可显著提高分类器的识别性能和稳健性。

Adaptive Construction of Hierarchical Classifiers Basing on the Confusion Relationship among Multiple Patterns

The determination of the hierarchical relationship and the objective patterns of sub - classifiers is the primary difficulty in the construction of a hierarchical classifier. To solve this problem effectively, firstly, the confusion relationship between patterns has been defined to describe the interweaving effects of patterns in the judgment domain. Then a measurement of this relationship has been proposed by utilizing the confusion matrix. Abiding by the Fisher Principle, a Multi - Patterns'Confusion Relationship Analysis Machine (MPCRAM), which integrates the supervised and the unsupervised methods for pattern recombination, has been designed to adaptively construct the structure of a hierarchical classifier. Various data scenarios and schemes have been used to compare the hierarchical structures generated with this method and those generated in conventional ways. The results have testified that the proposed method was effective and practical, and it could prominently improve the performance and robustness of a hierarchical classifier.