Improved YOLOv11n-Based PPE and Tool Object Detection for Power-Construction Safety Monitoring

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Aiming at the challenges of detecting personal protective equipment (PPE) and tools in power-construction scenes, including missed small objects, confusion between similar objects, inaccurate localization of pose-related objects, reduced robustness in complex backgrounds, and edge-device deployment constraints, this paper proposes an improved YOLOv11n object-detection model. The model embeds ECA and SGE attention mechanisms, replaces the baseline SPPF block with SimSPPF, and introduces the MPDIoU loss function. The resulting detector identifies seven object classes (helmet, person, insulating gloves, safety belt, operating rod, voltage tester, and work uniform); it does not directly classify violation behaviors. After integration of all modules, mAP@0.5 reaches 84.7% and mAP@0.5:0.95 reaches 56.6% on the power-construction dataset. The detected PPE and tool objects can serve as inputs to a subsequent rule layer for safety-violation judgment.