<p>As insulators must support conductors and provide insulation, their defects directly threaten the safety of power systems. Traditional manual inspections are inefficient, labor-intensive, and highly susceptible to weather conditions, with difficulty considerably increasing during under severe events such as heavy snow or dense fog. Although inspections using unmanned aerial vehicles offer a safe alternative through high-definition cameras and algorithms, adverse weather conditions degrade the imaging quality and blurs target features. Moreover, overlapping insulators and unclear contours can lead to background false positives or foreground false negatives. To address these problems, we propose the You Only Look Once (YOLO)-GHS network with three key optimizations. (1) YOLO-GHS incorporates the Hierarchical Graph Network (HGNet) V2 to enhance feature extraction, considerably improving object recognition in low-quality images. (2) To address the increased computational burden and parameter count introduced by the HGNetV2, YOLO-GHS employs ghost convolutions to reduce the computational load and optimize resource utilization. (3) YOLO-GHS introduces a novel spatial enhancement and attention module as a detection head, integrating spatial attention mechanisms with detail enhancement to precisely capture target features in overlapping regions, thereby suppressing background false positives and foreground false negatives. Experimental results demonstrate that YOLO-GHS outperforms the baseline YOLOv11 on the Insulator Defect Image Dataset, achieving improvements of 5.5%, 4.5%, 4.9%, and 5% in precision, recall, mean average precision, and F1-score, respectively. Hence, YOLO-GHS can accurately detect insulator defects in complex environments, thereby contributing to enhanced safety and reliability in power systems.</p>