Transmission line bird hazard detection and tracking based on LBIT-YOLO

This paper proposes a lightweight detection-tracking integrated model LBIT-YOLO to solve the engineering problems in intelligent bird hazard monitoring for transmission lines. These problems include insufficient multi-category bird recognition, weak multi-target tracking, poor lightweight adaptability for edge devices, high false-triggering rate of bird repellent equipment and difficult field deployment. Based on YOLOv11, the model adopts PConv, SimAM attention and PIoU loss to improve detection performance, and integrates Bytetrack to realize continuous multi-bird tracking. Experimental results show that the model achieves 76.3% mAP@50 and 59.7% mAP@50–95, with 5.2 and 8.9% points higher than the original YOLOv11. The IDF1 score of multi-object tracking reaches 80.073%, increased by 15.458% points. With 5.18 MB weight and 39.9 FPS edge inference speed, the model can be deployed on Jetson Xavier NX terminals efficiently. Field tests verify that the model greatly reduces the false actuation rate of bird repellent systems and meets the engineering requirements of transmission line operation. It provides a practical solution for precise bird hazard prevention and control of power transmission infrastructure.