Wear-YOLO: Research on Detection Methods of Safety Equipment for Power Personnel in Substations

Abstract

Targeting the target detection algorithm's limited generalization and accuracy for safety gear such as insulated gloves and helmets, and insulating shoes of traditional substation electric personnel, especially for the difficulty of detecting whether to wear insulating gloves or not, an improved YOLOv8 detection algorithm Wear-YOLO for substation power personnel safety equipment is proposed. To enhance comprehension of the contextual details of intricate settings, the C2f (CSP bottleneck with two convolutions) module of the Backbone part of YOLOv8 is replaced with the MobileViTv3 module that integrates the Transformer structure to capture long-distance dependencies and contextual information and combines it with local information. Fusion is done to enhance the feature extraction performance of the model in substation settings. In addition, a tiny target detection layer is added to improve the network's ability to extract shallow semantic information, which in turn improves the model's detection accuracy for small targets that are not wearing insulating gloves. This is done in order to maximize the impact of small target identification. A dynamic non-monotonic focusing method is proposed to help the model focus more on ordinary quality anchor boxes, boosting the model's flexibility to complicated conditions and detection accuracy. WIoUv3 is utilized to improve the bounding box regression loss function. According to the experimental results, the average detection accuracy is 92.1 %, the detection accuracy when wearing a helmet is 96.8%, the detection accuracy when insulating gloves are worn is 92.1%, and the detection accuracy when insulating shoes are worn is 93.0%. It is more accurate and resilient at detecting targets than the traditional target detection methods Faster R-CNN, SSD, RetinaNet, and YOLOv5.

Topics & Concepts