Research on oilfield worker violation detection algorithm based on improved YOLOv11

Oil serves as the cornerstone of national energy security. Violations such as workers failing to wear safety helmets, smoking, or using mobile phones during oilfield operations can easily trigger safety incidents. Existing detection methods exhibit substantial blind spots, degraded accuracy under low illumination, and difficulty in achieving a balance between precision and real-time efficiency. This paper proposes an enhanced YOLOv11-based model, termed OWDYOLO. The model incorporates an SPPF-LSKA module into its backbone network to strengthen small object feature extraction, employs an MAFPN structure at the neck to mitigate cross-scale feature mismatch issues, and integrates a DyHead module at the detection head to optimize target localization and classification performance. Experimental results demonstrate that the model achieves 86.9% precision, 81.4% recall, and 87.3% mAP@0.5, representing respective improvements of 2.7%, 1.3%, and 2.4%. Its accuracy outperforms seven mainstream algorithms including RT-DETR and YOLOv8. With 3.5 million parameters and 8.4 GFLOPs—only 8.2% and 6.7% of RT-DETR's respective values—it meets real-time detection demands for oilfield edge devices.