摘要
为了设计基于BP神经网络的字符识别系统模型并训练其参数,能够以较低硬件代价工程实现该系统,作者利用Matlab建立了一个基于BP神经网络的字符识别系统模型,对5000个样本进行了标记,并利用这些样本对模型进行了训练和验证,识别率达到了85.20%.同时,利用FPGA及Verilog硬件描述语言设计了该系统的神经元硬件电路,效果与Matlab的仿真数据一致.利用Altera公司的FPGA芯片实际综合下载了神经元及相关系统,能够实现对字符图像文件数据的计算.
The authors used Matlab to set up a character recognition system model based on BP neural network. As a result, its parameters can be trained and the system can be implemented at a low engineering hardware cost. We tagged 5000 samples which are used for model training and validation. The recognition rate of the method we propose reaches 85.20%. Meanwhile we use FPGA and Verilog HDL to design the neuron hardware circuit of the system, and the effect is consistent with the Matlab simulation data. The neuron circuit and related system is downloaded to a FPGA chip of Altera so that the data of character image file can be calculated.
引文
[1]张军阳,王慧丽,郭阳,等.深度学习相关研究综述[J].计算机应用研究,2018(07):1-12.
[2]吴忠,朱国龙,黄葛峰,等.基于图像识别技术的手写数字识别方法[J].计算机技术与发展,2011,21(12):48-51.
[3]唐忠,谢涛.Matlab神经网络工具NNTool的应用与仿真[J].计算机与现代化,2012(12):44-47,54.
[4]林祥金,张志利,朱智.人工神经网络FPGA实现研究进展与发展趋势[J].控制工程,2007(S3):1-3.
[5]徐远芳,周旸,郑华.基于MATLAB的BP神经网络实现研究[J].微型电脑应用,2006(08):41-44,3.
[6]罗成汉.基于MATLAB神经网络工具箱的BP网络实现[J].计算机仿真,2004(05):109-111,115.
[7]杨海钢,孙嘉斌,王慰.FPGA器件设计技术发展综述[J].电子与信息学报,2010,32(03):714-727.