基于卷积神经网络的数字识别

Digital recognition based on convolutional neural network

为了解决银行、邮局等场合的实时数字识别问题,提出了一种优化的卷积神经网络(Convolutionnal Neural Network,CNN)数字识别方法。以Lenet-5模型为基础改进了卷积神经网络结构并推导了改进后的前向和反向传播算法,将改进的卷积神经网络在手写、印刷数字组合数据库上进行测试,分析了不同样本数量、训练迭代次数等参数对识别准确率的影响,并与传统算法进行比较分析。结果表明改进后的CNN结构简单,处理速度快,识别准确率高,具有良好的鲁棒性和泛化性,识别性能明显高于传统网络结构。

In order to solve the problem of real-time digital identification in banks,post offices and other places,an optimized Convolutionnal Neural Network（CNN） digital recognition method is proposed.Based on the Lenet-5 model,the structure of the convolutional neural network is improved and the improved forward and backward propagation algorithms are derived,the improved convolution neural network is tested in the handwritten and printed digital combination database,and analyzes the influence of different training sample number,iteration number and other parameters on the accuracy of the identification,and compare and analyze with the traditional algorithm.The results show that the improved CNN has simple structure,fast processing speed,high recognition accuracy,good robustness and generalization,and the recognition performance is obviously higher than traditional network structure.