论文部分内容阅读
传统的模式识别方法主要存在两个方面的不足:集中式控制;依赖于特征抽取和选择的质量.本文的研究是在分析自上而下模式识别方法和自下而上模式识别方法的特性和优缺点基础上,基于多agent理论和技术,提出了将两种不同类型的方法集成的模式识别框架APRF(Agent-Based Pattern Recognition Frame):先用自下而上定量计算方法对模式建模,使其有利于分类;再用自上而下的定性分析方法对模式涌现.目的是让计算机模式识别更符合人的认知过程,APRF的基本思想是:模式识别=模式建模+模式涌现.
The traditional pattern recognition method has two main problems: centralized control and depends on the quality of feature extraction and selection.This paper is based on the analysis of the characteristics of the top-down pattern recognition method and the bottom-up pattern recognition method and Based on multi-agent theory and technology, this paper proposes a pattern recognition framework APRF (Agent-Based Pattern Recognition Frame) that integrates two different types of methods: First, the model is modeled by a bottom-up quantitative calculation method, Making it conducive to classification.And then using the top-down qualitative analysis method to emerge from the model.The purpose is to make computer pattern recognition more in line with human cognitive process, the basic idea of APRF is: Pattern Recognition = pattern modeling + pattern emerge.