基于Swin Transformer和CNN的汉字书法教学系统  

Chinese Character Calligraphy Teaching System Based on Swin Transformer and CNN

作　　者：林粤伟[1] 张通 宋丹 梁汇鑫 薛克程 LIN Yue-wei;ZHANG Tong;SONG Dan;LIANG Hui-xin;XUE Ke-cheng(College of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266061,China;Hisense Visual Technology Co.,Ltd.,Qingdao 266001,China;Shandong Xinhua Health Technology Co.,Ltd.,Zibo 255100,China)

机构地区：[1]青岛科技大学信息科学技术学院,青岛266061 [2]海信视像科技股份有限公司,青岛266001 [3]山东新华健康科技有限公司,淄博255100

出　　处：《青岛大学学报（自然科学版）》2024年第1期45-51,共7页Journal of Qingdao University(Natural Science Edition)

基　　金：青岛科技大学公派访学项目资助;2022年青岛科技大学教学改革研究面上项目(批准号:2022MS045)资助;2023年国家级大学生创新创业训练计划项目(批准号:202310426214)资助。

摘　　要：针对日益增长的汉字书法学习需求,将滑动窗口自注意力(Swin Transformer,ST)模型和卷积神经网络(Convolutional Neural Network,CNN)模型相结合,提出手写体汉字识别ST-CNN模型,进而开发了汉字书法教学系统。实测结果表明,ST-CNN模型识别准确率约为91.6%,较传统的ST模型提升了约0.5个百分点,较传统的CNN模型与ST模型,在收敛速度上分别提升了约10和30个百分点,开发的汉字书法教学系统性能良好。In response to the growing demand for Chinese calligraphy learning,a model combining the Swin Transformer(ST)and Convolutional Neural Network(CNN)was proposed for handwritten Chinese character recognition,subsequently leading to the development of a Chinese character calligraphy teaching system.The system employed an ST-CNN model for handwriting recognition and classification.The experimental results show that the recognition accuracy of the proposed ST-CNN model is around 91.6%,which has a 0.5 percentage points improvement over the traditional ST model.Moreover,the convergence speed of ST-CNN has been improved by about 10 and 30 percentage points respectively compared with traditional CNN and ST models.The developed calligraphy teaching system demonstrates good stability and performance.

关 键 词：深度学习 滑动窗口自注意力模型 卷积神经网络 手写体汉字识别 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]