基于深度学习的数字识别方法研究  

Research on Digital Recognition Method Based on Deep Learning

作　　者：唐子清 姚俭[1] TANG Zi-qing;YAO Jian(School of Management,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《软件导刊》2020年第9期228-232,共5页Software Guide

摘　　要：近年来,手写数字识别是计算机视觉与模式识别中一个广受关注的问题。该问题的主要挑战是如何设计一种有效方法,以识别用户通过数字设备提交的手写数字。目前,深度学习算法在计算机视觉领域非常流行,被用于处理诸如图像分类、自然语言处理及语音识别等问题。以几种深度学习常见算法,包括线性感知器、卷积神经网络、循环神经网络、长短时记忆网络等为研究对象,分析其在手写数字识别方面的优缺点,并引入Google第二代人工智能系统TensorFlow,对比相同算法在不同框架下的识别速度及准确率。实验结果表明,几类深度学习算法都能明显提高识别准确率,且在训练数据集时不会损耗过多计算资源。In recent years,handwritten numeral recognition is a widely concerned problem in computer vision and pattern recognition.The main challenge of this problem is to design an effective method to recognize handwritten numbers submitted by users through digital devices. At present,deep learning algorithm is very popular in computer vision,which is used to deal with important problems such as image classification,natural language processing and speech recognition. In this paper,several common algorithms of deep learning,including linear perceptron,convolutional neural network,cyclic neural network and short-term memory network are studied.Their advantages and disadvantages in handwritten numeral recognition are analyzed,and Tensorflow,the second generation of Google artificial intelligence system,is introduced. Then the recognition speed and accuracy of the same algorithm in different frameworks are compared. The results show that these deep learning algorithms can obviously improve the recognition rate,and do not consume too much computing resources when training data sets.

关 键 词：深度学习 手写数字识别 TensorFlow 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]