Improved the Detection Algorithm of Safety Helmet Wearing Based on YOLOv8

To settle challenges with low detection accuracy of small and strong objects, as well as complicated settings in practical construction scenarios like tunnels, coal mining operations, and building locations, an improved YOLOv8 safety helmet wearing detection algorithm has been established. To begin, the transformer part is the central component of the system. It uses the contextual knowledge stored in the feature diagram to help the model learn on its own and improve the linking of global feature information. Additionally, the bidirectional feature pyramid network (BiFPN) is employed to integrate characteristic data across various dimensions. Finally, important nodes in the network have the 3D weighted attention mechanism SimAM installed. This gets rid of unnecessary background details and focuses more on the unique features of the hard helmet. The experimental findings reveal that on the SHWD dataset, the revised algorithm's mAP (IOU = 0.5) increased by 3.8% to 95%, and the detection speed reached 127 FPS. In crowded, complex, and tiny object instances, detection increased significantly. It meets precision and real-time safety helmet detection standards and introduces a new solution for challenging construction situations.