摘要
学生课堂行为表现是课堂教学评价的重要组成部分,而进行学生课堂行为识别对课堂教学评价有重要意义。文章提出了基于人体骨架和深度学习的学生课堂行为识别方法,即通过提取学生行为图像的人体骨架关键信息,结合一个10层的深度卷积神经网络(CNN-10)来识别学生的课堂行为。为验证此方法的有效性,文章使用CNN-10和学生课堂行为识别方法,在学生课堂行为数据集上进行了对比实验。实验结果表明,学生课堂行为识别方法可有效排除学生体态、着装、教室背景等无关信息的干扰,突出关键有效信息,具有更高的识别准确率与泛化能力。使用基于人体骨架和深度学习的学生课堂行为识别方法识别学生典型的课堂行为,能及时、有效地反映学生的学习状态,并帮助教师精准掌握学生的课堂学情,从而助力智能化课堂教学。
The performance of student classroom behavior is an important part of classroom teaching evaluation,and conducting the student classroom behaviors recognition is of great significance to classroom teaching evaluation.In this paper,the recognition method of student classroom behaviors based on human skeleton and deep learning was proposed to recognize student classroom behaviors,by extracting the key information of human skeletons of student behavior images and combining with the a ten layer deep convolutional neural network(CNN-10).In order to verify the effectiveness of this method,the CNN-10 and the student classroom behavior recognition method were used to conduct comparative experiments on the dataset of student classroom behaviors.The experiment results showed that the student classroom behavior recognition method could effectively eliminate the interference of irrelevant information such as students’posture,dressing and classroom background,highlight the key effective information,and have higher recognition accuracy and generalization ability.Using the student classroom behavior recognition method based on human skeleton and deep learning to recognize students’typical classroom behaviors can reflect students’learning statuses timely and effectively,and help teachers accurately grasp students’classroom learning situations,thereby facilitating intelligent classroom teaching.
作者
何秀玲
杨凡
陈增照
方静
李洋洋
HE Xiu-ling;YANG Fan;CHEN Zeng-zhao;FANG Jing;LI Yang-yang(National Engineering Laboratory for Educational Big Data,Central China Normal University,Wuhan,Hubei,China 430079)
出处
《现代教育技术》
CSSCI
北大核心
2020年第11期105-112,共8页
Modern Educational Technology
基金
教育部人文社会科学研究规划基金项目“智慧教室环境下课堂交互有效性量化研究”(项目编号:17YJA880030)
华中师范大学中央高校基本科研业务费“高阶思维发展视角下的智慧课堂教学有效性评价研究”(项目编号:CCNU20ZT018)资助。
关键词
课堂行为识别
人体骨架信息
深度学习
卷积神经网络
数据集
classroom behavior recognition
human skeleton information
deep learning
convolutional neural network
dataset
