摘要
应用神经网络的方法,设计了一个432-26-36结构的三层BP训练网络,来识别具有一定噪声的26个大写英文字母及10个阿拉伯数字,并考察了网络初始参数对于收敛速度的影响,通过多次对试验结果的比较选取了最佳隐层节点个数。系统用MatLab进行仿真实现,试验结果表明该系统能够对具有一定噪声的大写英文字母和数字进行非常好的识别。
Applies the neural network to design a tri-level Back-Propagation network with a 432-26- 36 structure, and uses this network to recognize 26 upper case letters in English and 10 Arabic numerals with some noise, at the same time, tests the influence of the network＇s initial parameters to the speed of convergence, through comparing the results of many trials, chooses the best number of hidden nodes. This system uses MatLab to simulate, and the results of trials shows it can recognize the upper case letters in English and the numbers efficiently.
出处
《现代计算机》
2008年第10期59-63,共5页
Modern Computer
关键词
英文字母识别
数字识别
BP网络
MATLAB
噪声
English Letters Recognition
Numbers Recognition
Back-Propagation Network
MatLab
Noise
