摘要
当前卷积神经网络应用于手写数字的识别已成为研究的热点之一。本文在Matlab环境下输入手写数字图片,然后对图片进行灰度化、二值化、反色、去噪、分割和大小归一化预处理,通过卷积神经网络经典模型LeNet-5,对比3种数据集:MNIST数据集、MNIST数据集训练+自建数据集调精和自建数据集训练卷积神经网络的实际识别效果,选择自建的数据集进行卷积神经网络训练,在训练好的卷积神经网络中手写体数字图片取得了较好的识别效果。
Now,the application of Convolutional Neural Network in handwritten number recognition has become one of the research hotspots. In this paper,handwritten digital images are inputed in Matlab environment,and the images are preprocessed by graying,binarization,inverse color,de-noising,segmentation and size normalization. The classical model of Convolutional Neural Network Lenet-5 is used,and the comparison is provided of the actual recognition results of three kinds of data sets:MNIST training set,MNIST training set + own training and own training set. The proposed training set is selected for Convolutional Neural Network training,and for handwritten number pictures,good recognition results have been achieved in the trained Convolutional Neural Network.
作者
吕红
LV Hong(School of Information and Electrical Engineering,Xuzhou College of Industrial Technology,Xuzhou Jiangsu221000,China)
出处
《智能计算机与应用》
2019年第2期54-56,62,共4页
Intelligent Computer and Applications
关键词
手写体数字
深度学习
卷积神经网络
handwritten number
deep learning
Convolutional Neural Network
