文献类型：期刊文献

中文题名：基于人工智能的手写数字识别方法研究

英文题名：Research on Method of Handwritten Digit Recognition Based on Artificial Intelligence

作者：黄明春[1];田秀云[1];谢玉萍[1];王文华[1];谢钦[1];师文庆[1]

机构：[1]广东海洋大学电子与信息工程学院,广东湛江524088

年份：2023

卷号：52

期号：4

起止页码：185

中文期刊名：机电工程技术

外文期刊名：Mechanical & Electrical Engineering Technology

基金：2020年湛江市非资助科技攻关计划项目(编号:20092510540492);2021年广东海洋大学教育教学改革项目(编号:010201112104)。

语种：中文

中文关键词：手写数字识别;LeNet-5;ResNet残差网络

外文关键词：handwritten digital recognition;LeNet-5;ResNet residual network

中文摘要：手写数字识别属于图像分类问题。因个体手写数字的差异,传统的图像分类方法实现快速有效识别的难度相对较大。随着人工智能和计算机硬件技术的快速发展,基于深度学习卷积网络的手写数字识别逐渐成为研究热点。使用PyTorch搭建了经典的网络模型LeNet-5和改进的ResNet18模型进行手写数字识别。采用交叉熵损失函数和Adam优化算法,并设置学习率为0.001,在MNIST数据集上进行了训练和测试,鉴于ResNet18比LeNet-5网络结构深,在训练时花费的时间比LeNet-5多。经过100个Epoch后,使用LeNet-5模型在测试集上准确率达到了99.18%,使用ResNet18卷积模型的准确率高达99.55%,可以识别自制的手写数字,为人工智能识别系统的发展提供了一定的参考价值。

外文摘要：Handwritten digit recognition is an image classification problem.Due to the differences of individual handwritten digits,it is relatively difficult for traditional image classification methods to achieve fast and effective recognition.With the rapid development of artificial intelligence and computer hardware technology,handwritten digit recognition based on deep learning convolutional networks gradually becomes a research hotspot.The classical network model LeNet-5 and the improved ResNet18 model were built using PyTorch for handwritten digit recognition.The cross-entropy loss function,Adam optimization algorithm,and learning rate of 0.001 were used to train and test on the MNIST dataset.Given that ResNet18 had a deeper network structure than LeNet-5,it took more time to train than LeNet-5.After 100 epochs,the accuracy reached 99.18%on the test set using the LeNet-5 model and 99.55%using the ResNet18 convolutional model.It can recognize homemade handwritten digits,providing some reference value for the development of artificial intelligence recognition systems.