YOLOv5 Based Student Engagement and Emotional States Detection in E-Classes

Abstract

The rapid expansion of E-learning environments has highlighted the critical issue of cyberbullying within digital classrooms. This study introduces a novel approach for early detection of cyberbullying by analyzing student engagement and emotional states in real time. Our SER-YOLO model fuses an advanced You Only Look Once version 5 (YOLOv5) with a Student Emotion Recognition system, enriched by sophisticated methodological improvements. It features Soft NMS to refine the Non-Maximum Suppression (NMS) process, embeds the Channel Attention (CA) module to augment the network's backbone, and employs Enhanced Intersection over Union (EIOU) for bounding box regression. This method proactively detects changes in student engagement and emotional states, providing an effective mechanism for the early detection and management of cyberbullying in E-learning environments.