YOLO-HBW: A Hybrid Weighting Strategy for Defect Object Detection in UAV Power Grid Inspection

UAV inspection is crucial for power grid maintenance but faces challenges like scale variations, complex backgrounds, and severe long-tail class distribution, where rare defects are often missed by conventional detectors. To address this, we propose a hierarchical weighting strategy for a lightweight detector. This approach utilizes frequency-based grouping and fuses group-level and class-level weights to balance training stability with sensitivity to minority classes. Experimental results on two power grid inspection datasets demonstrate that our method significantly improves tail category performance and overall accuracy with negligible inference overhead, offering a practical solution for robust defect detection.