郭俊奇, 吕嘉昊, 王汝涵, 熊青云, 张世峰, 胡康颖. 深度学习模型驱动的师生课堂行为识别[J]. 北京师范大学学报(自然科学版), 2021, 57(6): 905-912. DOI: 10.12202/j.0476-0301.2021207
引用本文: 郭俊奇, 吕嘉昊, 王汝涵, 熊青云, 张世峰, 胡康颖. 深度学习模型驱动的师生课堂行为识别[J]. 北京师范大学学报(自然科学版), 2021, 57(6): 905-912. DOI: 10.12202/j.0476-0301.2021207
GUO Junqi, LÜ Jiahao, WANG Ruhan, XIONG Qingyun, ZHANG Shifeng, HU Kangying. Classroom behavior recognition driven by deep learning model[J]. Journal of Beijing Normal University(Natural Science), 2021, 57(6): 905-912. DOI: 10.12202/j.0476-0301.2021207
Citation: GUO Junqi, LÜ Jiahao, WANG Ruhan, XIONG Qingyun, ZHANG Shifeng, HU Kangying. Classroom behavior recognition driven by deep learning model[J]. Journal of Beijing Normal University(Natural Science), 2021, 57(6): 905-912. DOI: 10.12202/j.0476-0301.2021207

深度学习模型驱动的师生课堂行为识别

Classroom behavior recognition driven by deep learning model

  • 摘要: 根据课堂教学场景设计了三维卷积神经网络(3D-convolutional neural network,3D-CNN),以动态性为主要特征,对教师进行课堂行为识别;提出了经过改进损失函数的YOLO-v5(you only look once version 5th)模型,并以多目标为主要特征,对学生进行课堂行为识别.2种模型均取得了较好的识别结果.为验证所选用模型的有效性,在所标注课堂行为数据集上进行了模型性能对比试验.试验结果表明:所选用模型在教育场景下课堂行为识别工作中展现了较好的性能;课堂行为的精准识别能够帮助教师和学生了解课堂学情,有助于推动智慧课堂的发展.

     

    Abstract: Classroom behavior of teachers and students is important for performance evaluation.Deepened integration of information technology and education necessitates updated classroom behavior analysis, particularly with introduction of artificial intelligence technology.Here computer vision technology was applied to identify behaviors in classroom videos.3D-CNN was used to identify teacher behaviors with dynamic characteristics.YOLO-v5 was used to identify student behaviors with multi-objectives.All data sets were collected and annotated.To verify effectiveness of model selected, comparative model performance was examined in the same data set.Our model showed good performance in classroom behavior recognition.Accurate identification of classroom behavior would help both teachers and students to understand classroom learning process and promote development of smart classroom.

     

/

返回文章
返回