Online Classroom Student Engagement Analysis using Enhanced YOLOv5

Abstract

The surge in online education has underscored the pressing issue of cyberbullying in virtual classrooms. This paper introduces an inventive method for early cyberbullying detection by analyzing students' engagement and emotional responses in online classrooms. The proposed SFER-YOLOv5 model, integrating Student Facial Expression Recognition with an enhanced YOLOv5 object detection model, incorporates transformative optimizations. These include Soft NMS for Non-Maximum Suppression, the integration of a Channel Attention (CA) module within the YOLOv5 architecture, and the use of Enhanced Intersection over Union (EIOU) as the bounding box regression loss function. This approach identifies diminished engagement and emotional irregularities, offering a proactive framework for mitigating cyberbullying in online classrooms.