Facial Detection Based on Deep Learning for Emotion Recognition of Medical Staff

Based on the large amount of sample information from deep learning to establish a face detection model, the emotions of medical staff are detected. To address the issue of insufficient feature information fusion during the deep learning face detection process, which leads to low detection accuracy for multi-scale faces, an improved CBAM-YOLOv5m (Convolutional Block Attention Module-You Only Look Once version 5 medium) face detection model is proposed. The YOLOv5m face detection model can accurately locate the face area in real time under various lighting conditions and complex backgrounds, and quickly detect the key areas of the face in videos. The improved CBAM-YOLOv5m can enhance the discrimination ability of multi-scale features. In complex scenarios such as occlusion and uneven lighting, the false positive rate is reduced by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$10-15 \%$</tex>. On the COCO dataset, mAP{@} 0.5:0.95 can be improved by 2-4%, and ‘mAP{@} 0.5 ’ can be improved by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3-6 \%$</tex>. The inference speed is increased from 12.4 ms to 14.9 ms. Through comparison with RetinaFace and YOLOv8 face detection, it is verified that CBAM-YOLOv5m not only accurately locates the face area in real time under various lighting conditions and complex backgrounds and quickly detects the key areas of the face in videos, but also can more accurately detect the target in images containing multiple faces, occlusions, or people with different postures, solving the problem of face target detection of medical staff in different scenarios.