PyTorch下基于CNN的手写数字识别及应用研究

Research on Handwritten Number Recognition and Application Based on CNN Under PyTorch

文章对卷积神经网络模型LeNet-5中的激活函数、下采样方式等进行改进,对训练参数进行调整,使改进后的模型手写数字识别准确率达到99.2%。使用PyTorch搭建模型,用MNIST数据集对模型进行训练,其后在自制数据集上进行测试,从识别准确率和训练速度等方面验证了模型的可靠性。借助TensorBoard监督整个网络模型的训练过程,指导对模型参数的优化调整。最后,将改进的网络模型服务于该校人工智能课程答卷分数的识别中,使手写分数得到准确识别。

In this paper,the activation function and downsampling method of convolutional neural network model LeNet-5 are improved,and the training parameters are adjusted,so that the recognition accuracy of handwritten numerals in the improved model can reach 99.2%.PyTorch is used to build the model,MNIST data set is used to train the model,and then the model is tested on the self-made data set,which verifies the reliability of the model in terms of recognition accuracy and training speed.Tensor Board is used to supervise the training process of the whole network model and guide the optimization and adjustment of the model parame⁃ters.Finally,the improved network model is applied to the recognition of the scores of the AI course in the university,so that the handwritten scores can be recognized accurately.