基于YOWO的课堂学习行为实时识别  

YOWO Based Real-time Recognition of Classroom Learning Behaviors

作　　者：谢伟[1,2] 陶亚平[1,2] 高洁 周东波[4] 王溦[1,2] XIE Wei;TAO Ya-ping;GAO Jie;ZHOU Dong-bo;WANG Wei(Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning,Central China Normal University,Wuhan,Hubei,China 430079;School of Computer,Central China Normal University,Wuhan,Hubei,China 430079;School of Information Management,Central China Normal University,Wuhan,Hubei,China 430079;Faculty of Artificial in Education,Central China Normal University,Wuhan,Hubei,China 430079)

机构地区：[1]华中师范大学人工智能与智慧学习湖北省重点实验室,湖北武汉430079 [2]华中师范大学计算机学院,湖北武汉430079 [3]华中师范大学信息管理学院,湖北武汉430079 [4]华中师范大学人工智能教育学部,湖北武汉430079

出　　处：《现代教育技术》2022年第6期107-114,共8页Modern Educational Technology

基　　金：“信息化与基础教育均衡发展省部共建协同创新”中心重点项目“基于视频分析的乡村学校‘专递课堂’学生注意力获取与绩效关系建模研究”(项目编号:xtzd2021-004);科技创新2030新一代人工智能重大项目“混合增强在线教育关键技术与系统研究”(项目编号:2020AAA0108804);中央高校基本科研业务费项目“基于视频人体动作识别的课堂学习行为建模与分析”(项目编号:CCNU20TS028);华中师范大学教学改革研究项目“基于课堂学习行为分析与统计的教学方法优化研究”(项目编号:202013)资助。

摘　　要：当前,课堂学习行为研究一般仅针对单帧图像分析学生的行为,而忽视了行为的连续性,无法利用视频信息准确刻画学生的课堂学习行为。为了解决这一问题,文章实时识别真实课堂环境中学生的学习行为,首先采集某重点大学多个真实课堂教学中的授课视频,归纳出7种典型的课堂行为;随后,将行为识别中的YOWO模型迁移至课堂学生的学习行为识别场景中;最后,通过与基于图片数据集的VGG16网络模型、基于视频数据集的YOLO-V2网络模型进行实验对比,验证了YOWO模型对课堂视频中的学生学习行为具有较高的行为识别准确率。上述研究表明,使用YOWO模型识别课堂中典型的学生行为,可以促进教师及时了解学生的学习状态,进而优化教学过程,并能为课堂教学的过程化评价提供依据。At present,classroom learning behavior generally only analyzes students’behavior in single-frame images,but ignores the continuity of behavior,and cannot accurately describe students’classroom learning behaviors using video information.In order to solve this problem,this paper identified students’learning behaviors in a real environment in real time.Firstly,lecture videos from multiple real classroom teaching in a key university were collected,and further 7 typical classroom behaviors were summarized.Subsequently,the YOWO(You Only Watch One)network model in behavior recognition was transferred to the scene of classroom students’learning behavior recognition.Finally,through experimental comparison with the VGG-16 network model based on the picture dataset and the YOLO-V2 network model based on the video dataset,it was verified that the YOWO network model had a high behavior recognition accuracy rate for students’learning behaviors in classroom videos.The above research showed that using the YOWO model to identify typical students’behaviors in classrooms could not only help teachers to understand students’learning status in a timely manner,further optimize the teaching process,but also provide a basis for the procedural evaluation of classroom teaching.

关 键 词：学习行为 行为识别 视频分析 深度学习 YOWO模型 

分 类 号：G40-057[文化科学—教育学原理]