PD-YOLOv11: A power distribution enabled YOLOv11 algorithm for power transmission tower component detection in UAV inspection
Liangshuai Liu, Lingming Meng, An Li, Yakun Lv, Baijie Zhao
Abstract
Power distribution tower component detection in UAV inspections presents significant challenges, including multi-scale detection, small component localization, and overlapping target issues. To address these, we propose PD-YOLOv11, an enhanced object detection model that integrates key innovations: the C3K2_Sc backbone, CARAFE neck, and FASFFHead+Focaler-IOU detection head. These innovations optimize feature extraction and fusion for both large and small components, improving detection accuracy across a variety of scenarios. We evaluate PD-YOLOv11 on the combined NWPU VHR-10 and InsPLAD datasets, achieving an [email protected] of 0.823 and an [email protected]:0.95 of 0.657. These results significantly outperform existing models, including YOLOv11 and Faster R-CNN. This study highlights PD-YOLOv11’s potential for UAV-based power distribution tower component detection, demonstrating its superior accuracy and robustness. Future work will focus on optimizing the model for lightweight deployment and expanding the dataset to enhance its performance in diverse real-world conditions.