Research on Object Detection Algorithm for Important Components and Their Defects of Overhead Transmission Lines Based on Improved YOLOv8s

Currently, the inspection work for overhead transmission lines mainly relies on manual inspection by workers using drones for fixed-point inspection, which is prone to false detection and missed detection, and is not conducive to continuous tracking of the status of overhead transmission lines. Therefore, it is necessary to study object detection algorithms suitable for power inspection. However, high-precision object detection models generally have large computation and parameter amounts, making it difficult to deploy them on edge-side devices with limited computing resources for real-time inspection tasks. To this end, this paper proposes a object detection model for important components and defects of overhead transmission lines based on improved YOLOv8s, taking into account both accuracy and model size. Experimental results show that on a self-built dataset, compared to YOLOv8s, the proposed method reduces the parameter amount by $\mathbf{3 3. 6 \%}$, slightly increases the computation amount by 1%, and improves the mAP50 by 1.7%, proving the accuracy and efficiency of the proposed method in object detection of insulators and their defects, hardware fittings, and foreign objects on overhead transmission lines. On the basis of completing the algorithm design, this paper also deploys the algorithm on the Jetson edge-side computing platform through Deepstream, which has certain practical value for power inspection tasks.