基于同源的同类事物连通本性的模式分类神经网络模型

摘要

Based on the connected character of homologue, A pattern classification neural network model that can guarantee the correct connecting path of homologue is presented.This model includes the connecting and sequential technology of homologue samples and the sequential learning and incremental learing algorithms of topology sturcture of improved forward masking neural network model for establishment of connecting weight.The common potential problems of original sequential learning forward masking neural network model and many traditional pattern recognition methods that individual local connecting path of homologue are cut off can be overcome by this model which therefore enhance the categorizing ability.Moreover, this model can carry out rapid in-creamental learning for new added samples and hence is able to improve the categorizing and expanding ability of this model in a short time, which shows its advantage for large scale pattern recognition.Experiments also indicates that the pattern recognition model based on connectedness of homologue gives rise to a high correct recognition rate.This paper is of great significance to improving many traditional pattern recognition methods.%根据同源的同类事物连通的本质特性,本文提出保同类事物正确连通通路的模式分类神经网络模型.该模型包括同源的同类事物样本连通连网排序技术、改进的前向掩蔽神经网络模型拓扑结构的连接权值排序学习算法和改进的增量学习算法.本模型解决了原来排序学习前向掩蔽神经网络模型和许多传统的模式识别方法存在的共同隐患——把同源的同类事物的个别局部连通通路割断,提高了分类能力.而且,该模型还能对新增样本进行快速增量学习,从而能够在较短的时间内提高该网络模型的分类推广能力,能够在大规模模式识别场合发挥其优势.实验结果表明基于同类事物连通本性的模式识别模型的正确识别率高.本文最大意义在于,用本文思想方法可以改进一些传统的模式识别方法.