SHWDPPNet: Safety Helmet Wearing Detection in Power Production Environments Based on YOLOv8

In power production environments, wearing safety helmets is critical to ensuring the personal safety of workers. However, the effectiveness of existing helmet-wearing detection methods is often compromised in complex industrial settings due to challenges such as small target size, cluttered backgrounds, and similar colors between helmets and surroundings. To address these issues, a helmet-wearing detection approach tailored for power production environments is proposed in this study, which integrates YOLOv8 with transfer learning to enhance detection accuracy and adaptability. The proposed approach first leverages a large-scale COCO dataset to pretrain the YOLOv8 model. Subsequently, the pretrained model is fine-tuned using a dataset of safety helmet images to obtain the final detection model. Empirical evaluation results validate that the proposed approach is effective. For the “Wear_helmet” class, our method achieves an average precision (AP) of 49.9, outperforming the non-transfer learning YOLOv8 baseline by 5.4 AP. This demonstrates its potential as a novel solution for monitoring helmet-wearing compliance in complex power production environments.