基于CNN的高精度手写体数字识别

High Accurate Handwritten Number Recognition Based on CNN Network

近年来,人工智能和机器学习已成为国内外学者重要研究的领域。基于此,笔者探讨基于卷积神经网络(Convolutional Neural Networks,CNN)的高精度手写体数字识别,并将卷积内核大小、滤波器的数量、池化层的种类以及优化器的类型作为变量,通过改变这些变量做对比实验,检测模型训练和测试的准确率。实验结果表明,当卷积内核设置为3×3,三层卷积层滤波器的数量分别为32、64、128,使用最大池化层,选择Adam作为模型优化器的情况下,网络能够达到100%的训练准确率和99.55%的测试准确率,实现了高精度的手写体数字识别。

In recent years,artificial intelligence and machine learning have become important areas of concern for scholars at home and abroad.Based on this,the author discusses the high-precision handwritten digit recognition based on Convolutional Neural Networks(CNN),and takes the convolution kernel size,the number of filters,the type of pooling layer and the type of optimizer as variables.By changing these variables to do comparative experiments,the accuracy of model training and testing is tested.The experimental results indicate that the convolutional neural network model proposed in this paper can achieve 100% training accuracy and 99.55% validation accuracy under the condition that the convolution kernel is set as 3×3,the number of three-layer convolutional layer filters is 32,64 and 128 respectively,the max pooling layer is used and also Adam is selected as the model optimizer.