基于TensorFlow手写体数字识别系统的研究  

Research on handwritten digit recognition system based on Tensorflow

作　　者：李建新 LI Jianxin(Dept of Computer Science and Engineering,Guangzhou College of Technology and Business,Guangzhou 528138,China)

机构地区：[1]广州工商学院计算机科学与工程系,广州528138

出　　处：《智能计算机与应用》2020年第11期144-147,共4页Intelligent Computer and Applications

摘　　要：手写体数字识别是人工智能识别系统中的重要组成部分,本文基于TensorFlow深度学习模型,完成了手写体数字识别及应用。首先建立TensorFlow深度学习模型,分析了卷积神经网络(CNN)模型结构及Softmax模型结构,对手写体数据集M NIST中的60 000个样本进行深度学习,并对10 000个样本的测试进行对比。通过深度学习得到手写体数字的识别模型参数,进而使用模型参数识别用户手写体数字。实测数据验证,相对于传统的Softmax模型,基于TensorFlow深度学习的CNN模型准确率高达100%,提升了7%。该研究为人工智能数字识别系统应用提供了一定的科研价值。Handwritten numeral recognition is an important part of artificial intelligence recognition system.Handwritten numeral recognition and application are completed based on tensorflow deep learning model.Firstly,tensorflow deep learning model is established,convolutional neural network(CNN) model structure and softmax are analyzed,and then 60 000 samples of handwritten dataset MNIST are deeply learned,and 10 000 samples are processed Through deep learning,we can get the recognition model parameters of handwritten numbers,and use the model parameters to recognize the handwritten numbers of users.Compared with the traditional softmax model,the accuracy of CNN model based on tensorflow in-depth learning is up to 100%,7% higher.The research provides a certain research value for the application of AI digital recognition system.

关 键 词：机器学习 深度学习 卷积 TensorFlow 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]