A Lightweight Facial Recognition Model for Power Sites

Facial recognition technology, as a significant branch of artificial intelligence, is playing an increasingly important role in our lives. From smartphone unlocking and payment verification to public safety monitoring and access control systems, the applications of facial recognition are becoming more widespread. This paper proposes a lightweight facial recognition model applied to power security systems. The recognition algorithm combines the YOLOv5 and ResNet-101 networks. First, a face detection model is constructed using the YOLOv5 network. After extracting facial bounding boxes with this model, the detected faces are aligned, and then features are extracted using the ResNet-101 network. Finally, ArcFace is used as a metric function for facial feature comparison. The face detection model has an accuracy of 98%, with an [email protected] score exceeding 0.99. The proposed facial recognition model achieved an accuracy of 0.963 on the LFW dataset. By performing face detection and recognition on video streams of power plant workers, the system can provide data on worker identities, enabling monitoring of workers, preventing unauthorized access, controlling working hours, and providing effective data for subsequent site risk assessment.