Abstract: Aiming at the problems of poor adaptability and high false detection rate of the existing power facility staff's dress detection algorithm in actual operation scenarios, this paper proposes a dress detection model for power construction personnel combining Alphapose and ResNet. The model obtains the key point coordinates of people in the construction video through the Alphapose human pose estimation model, then adopts the appropriate cropping algorithm to cut out the required body area according to the coordinates, and finally feeds it into the improved ResNet classification model for dress detection and judgment work whether people wear safety helmets, work clothes and work boots. We use the data set constructed by ourselves to conduct experiments. The experimental results show that the method has high accuracy and low false detection rate, can meet the needs of real-time, and is suitable for application in the actual job site.