基于深度学习的学生课堂状态检测算法与应用  

Deep Learning Based on Student Classroom Status Detection Algorithm and Application

作　　者：史雨 辛宇 袁静[1] 欧阳群波[1] 杨德贺 袁国铭 

机构地区：[1]防灾科技学院,河北 廊坊 [2]国家自然灾害防治研究院,北京

出　　处：《人工智能与机器人研究》2021年第2期123-132,共10页Artificial Intelligence and Robotics Research

摘　　要：课堂教学评价是教学管理的重要组成部分之一,但依赖于督导教师完成该项工作的管理形式难以全面评价并反馈学生课堂学习状态。同时,我国高校的课堂监控视频数据被大量搁置未发挥作用。基于此,本文将传统教学管理与人工智能有机结合,提出学生课堂学习状态智能检测算法,采用K-means++聚类算法对目标候选框的个数和宽高比进行聚类分析,搭建双YOLO网络模型对课堂监控视频中学生的课堂行为进行分析,实时、精准地反馈学生的课堂学习状态,并对结果进行评分分级辅助督导教师进行课堂教学评价任务以提高教学管理效率。经过测试实验,本章提出的双YOLO网络模型的准确率为86.62%,且每帧教室监控图像的计算时间0.2 s。Classroom teaching evaluation is one of the important parts of teaching management, but it is diffi-cult to comprehensively evaluate and feedback students’ classroom learning status in the manage-ment form that relies on the supervisor to complete the work. At the same time, the classroom sur-veillance video data of colleges and universities in China has been largely shelved and has not played a role. Based on this, this article will combine traditional teaching management and artificial intelligence, students’ classroom learning intelligent detection algorithm is put forward, by the method of K-means++ clustering algorithm, the number and the aspect ratio of target candidate box for clustering analysis, build the double YOLO network model for monitoring video classroom of students classroom behavior analysis, real-time, accurate feedback on students’ classroom learning state, and the results are teachers’ classroom teaching evaluation rating auxiliary supervision task to improve the efficiency of teaching management. After testing and experiments, the accuracy rate of the TWO-STAGE YOLO network model proposed in this chapter is 86.62%, and the calculation time of each frame of classroom monitoring image is 0.2 s.

关 键 词：人工智能 学生课堂学习状态检测 双YOLO网络 K-means++ 

分 类 号：G63[文化科学—教育学]