基于改进遗传算法的神经网络手写体数字识别  

Neural network of handwritten digital recognition based on an improved genetic algorithm

作　　者：魏衡华[1] 彭飞[1] 

机构地区：[1]中国科学技术大学自动化系,安徽合肥230027

出　　处：《微型机与应用》2012年第18期57-59,共3页Microcomputer & Its Applications

摘　　要：BP网络作为人工神经网络的重要分支,已经广泛应用于手写数字识别。然而BP神经网络存在训练时间长、易陷入局部最小的问题。为了克服这些问题,提出了一种改进的遗传算法,并用该算法来优化神经网络的权值和阈值。最后,利用基于该算法的神经网络对大量USPS手写数字样本集进行训练。实验结果表明,该算法比单纯的BP算法具有更快的识别速率。As an important branch of artificial neural network, BP network has been widely applied to handwritten digital recognition. However, BP network has the problems of training for long time and going easily into the local minimum. In order to solve these problems, the paper presents an improved genetic algorithm to optimize the back-propagation neural network weights and thresholds. In the end, use it to train USPS handwritten digital sample. The experimental result show the proposed algorithm have faster recognition rate than the BP algorithm.

关 键 词：遗传算法 BP神经网络 数字识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]