基于卷积神经网络的手写数字识别技术研究  

Research on Handwritten Digit Recognition Technology Based on Convolutional Neural Network

作　　者：余国庆 杨燕婷 宗兆星 刘光宇 赵恩铭[1] 周豹 YU Guoqing;YANG Yanting;ZONG Zhaoxing;LIU Guangyu;ZHAO Enming;ZHOU Bao(School of Engineering,Dali University,Dali 671003,China;Key Laboratory of Marine Intelligent Equipment and System of Ministry of Education,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]大理大学工程学院,云南大理671003 [2]上海交通大学海洋智能装备与系统教育部重点实验室,上海200240

出　　处：《安徽电子信息职业技术学院学报》2024年第3期1-5,共5页Journal of Anhui Vocational College of Electronics & Information Technology

基　　金：海洋智能装备与系统教育部重点实验室开放基金项目“基于深度CNN的海底废弃物识别技术”(MIES-2023-02);云南省教育厅科学研究基金项目“基于Radon变换和特征空间决策的海洋目标检测”(2024Y855);云南省教育厅科学研究基金项目“基于水下声传播特性的洱海水质监测技术研究”(2024Y851)。

摘　　要：手写数字识别与我们的生活和工作息息相关,传统的人工判断手写数字方式需要耗费大量精力,并且存在准确性不高和时效性不能保证的弊端。为解决这一问题,采用卷积神经网络方法来进行手写数字识别。首先将MNIST数据集划分为训练集与测试集,其次在MATLAB上搭建训练LeNet-5与AlexNet两种卷积神经网络模型,之后将测试集导入模型,测试性能,最后再将验证集导入验证模型性能。实验结果表明,AlexNet模型对验证集识别率高于LeNet-5模型,且AlexNet模型的稳定性优于LeNet-5。Handwritten digit recognition is closely related to our daily life and work.Traditional manual judgment of handwritten digits requires a lot of effort,and there are also drawbacks such as low accuracy and inability to guarantee timeliness.To address this issue,convolutional neural network are used for handwritten digit recognition.Firstly,divide the MNIST dataset into training and testing sets.Secondly,build two types of convolutional neural network models,LeNet-5 and AlexNet,on MATLAB for training.Then,import the testing set into the model to test its performance.Finally,import the validation set to validate model performance.The experimental results show that the AlexNet model has a higher recognition rate for the validation set than the LeNet-5 model,and the stability of the AlexNet model is better than that of the LeNet-5 model.

关 键 词：LeNet-5卷积神经网络 手写数字识别 MNIST数据集 AlexNet卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]