Identification of single suspension point and single string insulators for cross crossing based on multi-source data

There is a situation where a single suspension point and single string insulator are used for the suspension of conductors at some crossing points of existing overhead transmission lines, which is prone to wire detachment and poses great safety hazards. This article proposes an improved YOLOv5 based method for identifying single suspension point and single string insulators at cross crossing points of overhead transmission lines. We devised a Lightweight Attention Mechanism Module (LAMCM) to bolster the network's feature extraction prowess and appended an additional prediction head atop the original YOLOv5's trio of prediction heads, aimed at enhancing the detection of targets across varying scales. Upon validating the augmented YOLOv5 algorithm on both the COCO dataset and a proprietary transmission line dataset, the outcomes revealed that our proposed algorithm surpasses numerous prevalent object detection methods. Specifically, on the COCO dataset, our algorithm achieved an impressive 79 FPS and 49.1% AP on a Tesla V100 platform. Meanwhile, on the self-constructed transmission line dataset, our algorithm registered a recognition recall rate of 96.8% and an accuracy rate of 94.3% for single suspension points and single string insulators, exhibiting enhancements of 1.7% and 1.8%, respectively, over the vanilla YOLOv5.