A Real-time Detection Method for Transmission-line Defects in UAV Inspection Based on a Lightweight Deep Learning Model

To address the difficulty of on-site real-time detection arising from the limited onboard computing power in UAV transmission-line inspection together with the large parameter size and high computational cost of existing deep-learning models, a lightweight method for detecting insulator self-explosion defects is proposed. With YOLOv8n as the baseline, a GhostNet lightweight backbone replaces the original backbone to reduce the parameter count and computation, and coordinate attention is introduced after the neck to compensate for the localization accuracy of small targets. Experiments on the CPLID dataset show that the method substantially compresses the model size and improves inference speed while maintaining detection accuracy, with an inference frame rate meeting real-time requirements, achieving a favourable balance among accuracy, size and speed and providing a feasible solution for on-board real-time detection of insulator self-explosion defects.