基于BP神经网络的数字识别研究  

Study of digit recognition based on BP neural network

作　　者：马耀名[1] 黄敏[1] 

机构地区：[1]辽宁工程技术大学电气与控制工程学院,葫芦岛125105

出　　处：《信息技术》2007年第4期87-88,91,共3页Information Technology

摘　　要：为了提高数字识别的准确性和可靠性,将BP神经网络应用于数字识别,并考察了网络结构对系统实时性、识别效果的影响以及初始权值对收敛速度的影响,选取了最佳的隐层节点数和权值初始化函数,实现了基于BP神经网络的数字识别系统。利用MATLAB进行仿真实验,实验结果表明该系统能够实现对数字的有效识别。In this paper, BP neural network is applied to digit recognition system in order to improve the accuracy and reliability of system. For fulfilling the real time and veracity demands of digit recognition, the structure of neural networks is researched. The influence of initial weights on convergent speed is also considered. Based on the successful selections of hidden layers and weight initialization functions, this paper realized a digit recognition system based on BP neural network. It evaluated the performance of system in MATLAB. Experimental results demonstrate that the method works well in recognizing the digit.

关 键 词：数字识别 神经网络 BP算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]