Defect detection of fasteners in transmission lines based on Real-ESRGAN and improved YOLOv11n

To address the challenge of detecting defective fasteners in transmission lines during UAV inspections, a new method for insulator fault detection is proposed. This method integrates Real-World Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) and an improved YOLOv11n. First, Real-ESRGAN is employed for super-resolution reconstruction of the dataset, enhancing dataset quality and effectively reducing interference from complex backgrounds. Then, an efficient vision transformer framework is used to replace the backbone of YOLOv11n, improving the model's feature extraction capabilities while accelerating processing speed during inference. Finally, lightweight optimization is applied to the YOLOv11n detection head, further speeding up model inference. Experimental results demonstrate that the proposed method achieves a mean average precision (mAP) of 91.4%, showcasing superior detection performance in complex backgrounds. By analyzing heatmaps, the differences in attention regions between the proposed algorithm and the original YOLOv11n are revealed, providing insights into the model's internal mechanisms.