摘要
本文从模拟人脑思维功能这一基本思想出发,提出了一种改进的多层神经网络学习算法,并用于自由手写字体数字的识别,同时也提出了独特的特征加权算法.模拟结果表明对于变化较大或倾斜的手写字体数字能实现较精确的识别.
An improved learning algorithm of multilayer neural networks based on simulation of human brains is provided for handwritten numeral recognition, and also presented is the weighted feature algorithm. The computer simulation shows that the neural networks can correctly recognize the inclined and irregular handwritten numerals.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
1991年第3期8-15,共8页
Journal of Southeast University：Natural Science Edition
关键词
神经网络
数字识别
手写体
network, feature vector / learning samples
