To address the requirements for high‐precision detection of transmission line defects by inspection drones in low‐light environments such as cloudy days and to overcome the problem of significant accuracy degradation in current defect detection algorithms under low‐light conditions, this paper uses YOLOv8 as the baseline algorithm. By introducing the PENet low‐light enhancement network, the Slim‐Neck lightweight neck network, and the Wise‐IoU loss function, it forms the low‐light line defect detection method LCDD‐YOLO. Simultaneously, to address the current lack of images in low‐light environments, this paper explores a low‐light image generation scheme based on the CycleGAN method and constructs the comprehensive transmission line defect dataset (TLCDD). Dataset comparison experiments verified the effectiveness of the TLCDD dataset constructed in this paper in improving the accuracy of algorithm defect detection in low light. Through ablation experiments, compared with the baseline YOLOv8 algorithm, LCDD‐YOLO achieves a 10.583% improvement in mAP@0.5 while incurring only a small increase of 2.9 GFLOPs and a slight decrease of 5 FPS. Furthermore, in comparative experiments, LCDD‐YOLO demonstrated the highest defect recognition accuracy in low‐light environments, proving its superior performance and its ability to meet the demand for detecting transmission line defects in low‐light environments.
Zhu et al. (Thu,) studied this question.