摘要
手写数字识别是模式识别中非常重要和关键的研究领域,并且具有一定的难度。每个人的手写习惯各不相同,而且手写数字识别很难建立精确的数学模型,使得手写数字的识别难度大大增加。针对手写数字难以准确识别的问题,将LeNet神经网络应用到其中,并对LeNet模型的输入层、卷积层、池化层及全连接层进行原理方面的剖析。基于TensorFlow的框架和Python语言进行了相关的程序编写与设计,以MNIST库作为数据集的来源,使用PyQt5设计GUI画板界面,最终实现对于手写数字的识别。实验结果表明,本系统的识别准确度和辨识度达到99.2%,可应用于发票手写数字识别、酒店登记手写身份证识别等生活实践中。
Handwritten digit recognition is a very important and key research field in pattern recognition,and has certain difficulty.Handwriting habits of each person are different,and handwriting digit recognition is difficult to establish accurate mathematical model,which makes handwriting digit recognition greatly difficult.Aiming at the problem that handwritten digits are difficult to be accurately identified,LeNet neural network is applied to it,and the input layer,convolution layer,pooling layer and fully connected layer of LeNet model are analyzed in principle.Based on TensorFlow framework and Python language,related programming and design,MNIST library as the source of the dataset,using PyQt5 design GUI sketchboard interface,finally realize the recognition of handwritten digits.The recognition accuracy and recognition degree of this system reached 99.2%,which can be applied to the invoice handwritten digit recognition,hotel registration handwritten ID identification and other life practices.
作者
任东悦
李名博
卫勇
REN Dongyue;LI Mingbo;WEI Yong(College of Engineering and Technology,Tianjin Agricultural University,Tianjin 300380,China)
出处
《传感器世界》
2023年第2期26-31,共6页
Sensor World
关键词
深度学习
卷积神经网络
手写数字识别
特征提取
可视化
deep learning
convolutional neural network
handwritten digit recognition
feature extraction
visualization
