Identification of bird’s nest hazard level of transmission line based on improved yolov5 and location constraints
Yang Wu*, Qunsheng Zeng, Peng Li, Wenqi Huang, Lingyu Liang and Jiajie Chen
Published: November 19, 2022
Abstract
Bird's nest is a common defect in transmission line, which seriously affects the safe and stable operation of the line. This paper presents a method of bird's nest hazard level identification based on improved yolov5 and location constraints, which solves the problem of bird's nest multiple identification and hazard level classification. We integrate GhostModule and efficient channel attention (ECA) to design a lightweight attention mechanism convolution module (LAMCM). The original yolov5 is improved by using LAMCM and adding a prediction head, which improves the detection ability of small targets and alleviates the negative impact of scale violence. We only identify the bird's nest on the panorama of Unmanned Aerial Vehicle (UAV) patrol, and classify the hazard level of the bird's nest according to the location constraints of the bird's nest and insulator. Experiments on coco dataset and self built transmission line dataset (TL) show that our algorithm is superior to other commonly used algorithms. On the COCO dataset, our algorithm achieved 49.1% AP at a real-time speed of ~79FPS on Tesla V100. On the TL dataset, the recognition effect of our algorithm on towers, insulators and bird nests has improved to varying degrees compared with the original yolov5. In particular, the recall rate of bird nest identification for three hazard levels has increased by more than 3%. The average recall rate of bird nest hazard level identification is 93%, and the average accuracy rate is 93.5%.
Keywords: bird's nest; transmission line; object detection; attention mechanism