Abstract: Based on the operation process monitoring requirements of workers on offshore platforms,taking the wearing detection of labor protection appliances of on-site workers as the study goal,the SE attention mechanism and Focaleriou loss function are introduced to optimize the YOLOv9 model so as to improve the feature extraction ability and detection accuracy of YOLOv9 model. The longitudinal comparison experiment before and after model optimization shows that the mAP@0.5 value of improved model is increased by 3.0% compared to the original model,which proves the detection accuracy advantage of the improved model. The horizontal comparison experiment with other commonly used models shows that the improved model has highest F1 scores on multiple targets,which indicates that the improved model has better generalization performance. The improved model achieves high-precision detection of on-site workers and their wearing of labor protection appliances.

Key words: offshore platform, YOLOv9 model, labor protection appliance, object detection, model optimization