Underground helmet detection algorithm based on improved YOLOv8s

In the process of coal mine underground operation, the safety helmet is the most direct and effective protective measure, and it is an important measure to ensure the safety of miners. The safety helmet belongs to small target detection, and the underground operation environment is complex, and dust and light noise have a great impact on camera detection. In order to solve the above problems, this study proposes a detection algorithm for underground safety helmets based on improved YOLOv8s, which is called PBSS-YOLOv8. The PBSS-YOLOv8 model first adds the small target detection layer P2 to improve the detection performance of small targets, and on this basis, the BiFPN network framework is introduced to make the information in the network can be transmitted in both directions, which further improves the capture performance of the helmet, and the original P5 detection layer is deleted, which greatly reduces the number of parameters and calculations of the model. Then, the SPD-Conv convolution module is added, and the non-step convolutional layer is used to reduce the situations that the redundant information of small targets is filtered and fine-grained information is lost. Finally, the lightweight attention mechanism SGE is added to reduce the influence of environment and noise on target feature extraction. Experimental results show that compared with YOLOv8s, the improved PBSS-YOLOv8 increases the Helmet AP by 1.1%, the No-Helmet AP by 3.5%, the mAP by 2.8%, and the parameter amount decreases by 2 M. The experimental results confirm that the improved model can effectively improve the problem of missed detection and false detection of small targets, and provide a guarantee for the safe operation of underground personnel.