Classroom Behavior Recognition Through Joint Improvement of Faster RCNN Algorithm Under the Construction of Smart Education

• Jianqiang Dai,
• Aimin Xie,
• Deying Yu,

Vol. 19, No. 2, February 28, 2025
10.3837/tiis.2025.02.009, Download Paper (Free):

• behavior analysis
• Classroom video
• Deep Learning
• Faster R-CNN network
• Teaching behavior recognition

Abstract

In the context of smart education, student and teacher behavior identification has become an important research task. To better advance the construction of smart education, the study first constructed a dataset of teacher-student behavior, which includes various behaviors of students and teachers in the classroom, such as raising hands, writing, and explaining. Secondly, the trap detection algorithm Faster R-CNN was used to detect students' human body areas, and the VGG16 model was used to optimize the feature map extraction part. Combined with the loss function, the algorithm was optimized to obtain the final improved algorithm. The experiment outcomes denote that the recognition rate of the research method has reached 91.25%. Compared with other methods, the effectiveness of the raised model has been significantly improved, with average accuracy, average recall, and average F1 score reaching 99.23%, 97.65%, and 90.24%, respectively. The above findings illustrate that the study can effectively identify the behavior of students and teachers in the classroom, and provide real-time teaching support, providing strong support for the implementation of smart education.

Statistics

Cite this article