Abstract: In view of a lack of automatic supervision functions in industrial production safety systems, this paper designs a personnel dress detection algorithm for operators in construction sites, docks, mines and other construction sites, which is based on the localization of human keypoints and the attention mechanism of image areas to ensure the safety by carrying out the standard test for uniform wearing. The method innovatively combines the human pose estimation to locate the dress area, proposes an attention mechanism based on the image areas to effectively represent the features,and decouples the complex dress detection task into object detection and image classification tasks. The proposed approach improves the efficiency and performance of personnel dress detection in industrial scenarios. The experiments on the MSCOCO anddress detection datasets in the customizedcoal mine scene show that the proposed model achieves excellent results in personnel localization and dress detection tasks (AP50 is the best on the MSCOCO dataset). It is robust and can be applied to a variety of complex environments.