In response to the problem of face recognition for dense crowds in school classrooms, the study first introduces a face detection algorithm based on improved YOLOv5. Subsequently, the FaceNet algorithm was introduced and improved to perform facial recognition on densely populated student populations in the classroom. The algorithm further recognizes faces based on the improved detection of YOLOv5 and matches them with the school's facial database to verify their identity. In the comparison of face detection losses, the research model has the lowest loss of 0.03 in occluded scenes, which is superior to similar models. At the same time, in the accuracy of face detection in different poses, the research model has the best performance with a detection accuracy of 0.982. Overall, the method proposed by the research institute has high recognition accuracy in facial recognition of dense populations, providing an effective method reference for the problem of dense facial recognition in modern school classrooms.