Research on Image Recognition Algorithms for Foreign Object Intrusion in Transmission Lines

With the increasing frequency of external damages to transmission lines, traditional manual monitoring methods consume significant human resources. This paper proposes a computer vision-based image recognition algorithm for foreign object intrusion in transmission lines. First, a transmission line recognition algorithm based on the instance segmentation model LaneNet is proposed, achieving an accuracy of 90.85%. Second, a foreign object detection algorithm based on the DenseBox model is introduced, with an accuracy of 92.25%. Finally, a monocular vision-based 3D object localization algorithm is proposed, with experimental results showing an error below 3.11%. The proposed algorithm offers an effective solution for the automation of transmission line monitoring, enhancing the automation and intelligence of power grid operation and maintenance.