Abstract:Fatigue state detection plays an important role in protecting the safety of electric power workers.Although deep learning has become an important method for facial fatigue detection, the accuracy of its detection results still needs to be improved.To solve this problem, a new fatigue status recognition system for electric power workers is proposed.The system uses face detection algorithmon the collected video images to determine the facial position of workers in power scenes, uses YOLOv4-tiny target detection algorithm to extract the visual features of eyes, establishes detection model to improve the discriminative ability of visual features, and uses PERCLOSE criterion to analyze blink frequency and combine yawn frequency and doze nod frequency to determine whether the workers in the scene are tired or not.The experimental results prove the effectiveness of the proposed fatigue state detection method based on deep learning.