Detection Model of Substation Personnel Safety Protection Tools Based on Improved YOLOv5

Abstract

Substation operation and maintenance personnel face the hazards of high voltage and strong current in their work. Therefore, it is crucial to accurately and promptly detect the condition of safety protection tools for these personnel. This paper proposes a detection model that enhances YOLOv5‘s substation personnel safety protection tool detection capabilities. The CBAM attention mechanism was integrated into the YOLOv5 model using this approach, while the feature fusion network PANet was replaced with the CB-BiFPN structure. Finally, DIoU was used instead of GIoU loss function to improve the convergence speed and detection accuracy of the model. Experimental results demonstrate that our improved YOLOv5 model achieves high accuracy and fast processing speed with a mean average precision (mAP) of 89.23% at 32 frames per second (FPS). Our proposed method outperforms other related object detection models such as Faster RCNN, SSD, and YOLOv5 in terms of performance metrics. This study improves safety protection tool detection accuracy and provides valuable insights for real-time safety management and control in substations.