基于Keras和卷积神经网络CNN的手写数字识别研究  

Research on handwritten digit recognition based on Keras and convolutional neural network CNN

作　　者：赵亚腾 孙钰 ZHAO Yateng;SUN Yu(Jinzhong College of Information,Jinzhong,Shanxi 030800,China)

机构地区：[1]晋中信息学院,山西晋中030800

出　　处：《计算机应用文摘》2023年第5期110-112,共3页Chinese Journal of Computer Application

摘　　要：科技及信息化的应用在生活中随处可见,但也会带来一些问题,如手写试卷的文字识别精准度太低等。文章在深度学习中利用卷积神经网络完成手写数字识别模型的训练。为了构建出合理有效的模型结构,需要使用卷积网络在卷积层中安排适量的卷积核(也叫过滤器),并结合训练的数据得以让模型学习反映不同的手写数字卷积核心的10个卷积权重,最后通过全连接层使用softmax函数给出与每个数字的概率相对应的数字地图的预测概率。卷积神经网络的应用是将手写数字图像转化为数字标签,在MINST数据集上实现数据识别。首先,对数据进行预处理,要对WNIST数据集中将近60000个训练数据和10000个测试数据进行分析计算。数据的组成结构分为图像(数字图像)和标签.(实数)两部分。用6000*28x28x的二维矩阵形式把数字识别特征值演示出来。其次,利用深度学习框架Keras的特性训练MINST数据集,最终生成高精度识别模型,可以使手写数字的识别精准度达到99%,识别效果和速率大大提高。The application of science and technology and informatzation can be seen everywhere in life,but it will also bring some problems,such as the poor accuracy of handwriting test paper character recognition.This paper will use convolutional neural network to complete the training of handwritten digit recognition model in depth learning.In order to build a reasonable and effective model structure,it is necessary to use the convolution network to arrange an appropriate amount of convolution cores(also called filters)in the convolution layer,and combine the training data to.enable the model to learn ten convolution weights that reflect different handwritten digital conwolution cores.Finally,through the full connection layer,the softmax function is used to give the prediction probability of the digital map corresponding to the probability of each number.Use 6000*28x28x two-dimensional matrix to demonstrate the characteristic value of digital recognition.Secondly,using the characteristics of the deep learning framework Keras to train the final high-precision recognition model generated from the MINST dataset can make the recognition accuracy of handwritten numerals reach 99%,and the recognition effect and speed are greatly improved.

关 键 词：卷积神经网络 MNIST 手写体数字识别 Keras 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]