Computer Engineering and Applications ›› 2007, Vol. 43 ›› Issue (13): 52-54.

• 学术探讨 • Previous Articles     Next Articles

Improvements of ART2 neutral network based on amplitude weight

  

  • Received:2006-09-15 Revised:1900-01-01 Online:2007-05-01 Published:2007-05-01

基于幅值分量的ART2神经网络的改进

顾民 葛良全   

  1. 成都理工大学核技术与自动化学院 成都理工大学
  • 通讯作者: 顾民

Abstract: Abstract: The measurement of similarity in ART2 neural network is something about the phases of pattern vectors during the processing of input and is limited of the same phase data with different amplitudes. Based on this, the article improve classical ART2 network by taking amplitude as character weight .The simulation verified that the improved ART2 network gain higher recognition rate than the classical one when grouping clustering data. The improved ART2 has definite value in data classify of nuclear radiation field.

Key words: ART2 network amplitude information phase information similarity

摘要: 摘要: ART2神经网络由于其算法结构中固有的归一化环节,丢失了幅度信息,其相似量度是一种模式相位信息的量度,存在“同相位不可分”的缺点。文章针对此不足,将样本的幅度作为样本特征分量的办法,对传统的ART2网络进行了改进。实验证明,改进后ART2网络在处理集群分布样本时,性能优于传统ART2网络,同时,改进的ART2网络在核辐射场数据处理分类中有一定的实用价值

关键词: ART2网络 幅度信息 相位信息 相似度