基于TensorFlow的CNN自由手写数字识别研究  

CNN Free Handwritten Numeral Recognition Research Based on Tensorflow

作　　者：徐颂民 江玉珍 XU Song-min;JIANG Yu-zhen(School of Continuing Education,South China University of Technology,Guangzhou 510000,China;School of Computer Infor⁃mation Engineering,Hanshan Normal University,Chaozhou 521000,China)

机构地区：[1]华南理工大学继续教育学院,广东广州510000 [2]韩山师范学院计算机与信息工程学院,广东潮州521000

出　　处：《电脑知识与技术》2020年第13期11-12,共2页Computer Knowledge and Technology

基　　金：韩山师范学院质量工程与教育教学改革项目(粤韩师教字〔2019〕67号)。

摘　　要：CNN卷积神经网络是图像识别和分类等领域的前沿研究方法。由于CNN模型训练效果与实际测试之间存在较大的差距,为提高自由手写数字的识别率,尝试使用TensorFlow搭构CNN网络模型,在完成MNIST数据集训练的基础上实现对自由手写数字的识别,并根据两种样本状态的差别和识别结果提出效果分析及改进方法,实验证明该改进方法获得明显效果。Convolution neural network is an advanced research method in the field of image recognition and classification.There is a big gap between the training effect and the actual test in the CNN model.To improve the recognition rate of free handwritten numer⁃al,a CNN numeral model based on tensorflow is constructed to perform recognition by having trained on the MNIST datasets.The effect analysis and improvement method are put forward according to the differences of two samples’status.Experiment shows the obviously improved result.

关 键 词：手写数字 卷积 神经网络 深度学习 训练本样 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]