Reliable power transmission is essential for uninterrupted electricity supply. Conductor defects can reduce system reliability and create operational risks if not detected early. Conventional inspection mainly depends on manual observation, which is labour intensive, time consuming and prone to human error. This study aims to develop an unmanned aerial vehicle (UAV)-assisted deep learning framework for detecting faults in power transmission conductors.
The proposed framework uses UAV-based image acquisition and YOLOv8s-based object detection. A total of 9,182 raw UAV images were initially collected using a UAV-mounted GoPro Hero7 Black camera. After removing blurred, redundant and near-duplicate images, 4,100 annotated conductor images were prepared for training, validation and testing. The dataset included two classes: Normal Conductor and Defective Conductor. RGB UAV images were used for YOLOv8s training, while thermal images were examined separately as complementary qualitative evidence for hotspot-related abnormalities.
The YOLOv8s detector achieved precision of 0.967, recall of 0.976, mAP@0.5 of 0.991 and mAP@0.5:0.95 of 0.815. The model also achieved an offline inference speed of approximately 59.17 FPS on an NVIDIA Tesla T4 GPU. These results indicate that the framework can accurately localize and classify normal and defective conductor regions under UAV-based outdoor inspection conditions.
This study presents a UAV-assisted YOLOv8s framework for conductor fault detection using a curated RGB image dataset with complementary thermal observations. The approach provides an accurate, efficient and safer alternative to conventional manual inspection.
