Classroom Attention Detection Using YOlOv8 and ResNet-based Facial Expression Recognition Model

This paper proposes a classroom attention detection model based on YOLOv8 and ResNet, aimed at evaluating students' engagement during class by analyzing facial expressions. The model consists of two main components: the YOLOv8 and ResNet classification models. YOLOv8 is used to localize facial regions in real time, and ResNet is used to recognize students' emotional states. Emotional states are categorized into seven facial expressions: "Surprise", "Fear", "Disgust", "Happy", "Sad", "Angry", and "Neutral". Positive expressions such as "Happy", "Surprise", and "Neutral" are classified as "Good" attention states, while negative expressions are classified as "Bad" attention states. Experimental results show that the proposed model achieves high precision and recall, with a precision of 0.88649, recall of 0.6654, mAP@50 of 0.7567, and mAP@50-95 of 0.41387. These results indicate that the model can effectively assess students' classroom attention, providing valuable feedback for teachers to monitor student engagement. In conclusion, this study demonstrates the potential of using facial expression recognition for classroom attention detection. Future work will focus on improving the model's real-time performance, adaptability across different environments, and integrating multimodal data for more accurate and personalized attention evaluation.