An Improved Yolov5s Algorithm for Emotion Detection

Affected by the regularity of the epidemic, many schools have launched online education. Although it is possible to monitor the status of students in class through cameras, teachers cannot observe the status of students in real time and adjust their teaching methods in time like traditional classes, so as to further improve the quality of online teaching. This study focuses on the facial features of students in class (such as neutral, happy, sad, surprised, disgusted, scared and angry), and improves the Yolov5 target detection algorithm to make it suitable for faster and more accurate emotion detection, so as to meet the real-time demand of detection. The original NMS used by the original Yolov5 is not friendly to prediction boxes with high consistency. Therefore, merge-YOLOV5 adds merge-NMS on the basis of the original Yolov5 to make prediction boxes more accurate. The experimental results show that the mAP@0.5:0.95 of the improved Yolov5 model is 0.5 percent higher than the original model, in which the feature recognition of happy and sad expressions which can reflect the classroom effect is the most improved.