摘要
在模式识别中,通常直接用神经网络来处理复杂的多类分类问题,其识别的误判率较大。该文基于任务分解与模块整合的思想,提出了一个模块化Kohonen神经网络(KTD)结构用于模式分类,给出了其学习方法并做了模拟仿真,模拟仿真表明KTD能够获得较高的识别率且误判率较小。
Usually,neural network is used directly for solving complicated and multiple-classify problems in pattern recognition.But this method brings higher misclassifation rate.This paper proposes a new modular Kohonen neural.network based on task decomposition and module combination for pattern recognition.It also give s the principle.Finally the paper performs simulation and the result of simulation showes that KTD has better recognition.
出处
《计算机工程与应用》
CSCD
北大核心
2002年第5期83-85,共3页
Computer Engineering and Applications
