YOLO-PowerSE: An Efficient Deep Learning Model for UAV-Based Power Cable Fault Detection

This paper presents YOLO-PowerSE, an efficient adaptation of YOLOv5 specifically designed for power cable fault detection from UAV imagery. Power transmission networks require regular inspection to prevent failures, with UAVs offering a promising alternative to traditional manual methods. However, detecting thin power cable structures and subtle fault features against complex backgrounds remains challenging even for state-of-the-art models. To address these issues, we integrate squeeze-and-excitation (SE) attention mechanisms into the YOLOv5 framework, enhancing the network's ability to focus on the most discriminative features critical for fault detection. The SE modules recalibrate channel-wise feature responses adaptively, highlighting important patterns while suppressing less useful ones. Experimental results on a custom dataset of UAV-captured power cable images demonstrate that YOLO-PowerSE achieves improved detection accuracy compared to the original YOLOv5 model. The greatest improvements are observed in detecting subtle fault types such as minor insulation damage and early-stage corrosion. This work offers a balanced solution that enhances detection performance for power cable fault detection while maintaining reasonable computational efficiency for UAV deployment.