RTF-SAW-YOLOv11: A Bolt Defect Detection Model for Power Transmission Lines Under Low-Light Conditions

To address the performance degradation of transmission line bolt defect detection under low-light conditions, we propose a unified model named RTF-SAW-YOLOv11, which integrates image enhancement and object detection into a single end-to-end framework. The model incorporates the Retinexformer (RTF) module for illumination compensation, noise suppression, and texture enhancement, and employs an improved YOLOv11-based SAW-YOLOv11 detector with Shallow Robust Feature Downsampling (SRFD) and Deep Robust Feature Downsampling (DRFD) to strengthen small-object feature extraction. An auxiliary detection branch (Aux Head) is introduced to enhance training stability, and the Wise IoU (WIoU) loss is applied to improve localization for irregular bolt targets. Experimental results on the low-light BoltData dataset show that RTF-SAW-YOLOv11 achieves 85.5% precision, 83.2% recall, and 88.0% mAP@0.5, outperforming the base RTF-YOLOv11 by 4.8%, 1.5%, and 2.7%, respectively. Compared with SAW-YOLOv11 alone, it achieves improvements of 11.1% in precision, 14.1% in recall, and 13.6% in mAP@0.5. RTF-SAW-YOLOv11 maintains real-time inference at 18.6 ms, with a total parameter size of 4.16 M, ensuring deployment feasibility. Generalization tests on the ExDark dataset confirm SAW-YOLOv11’s robustness. RTF-SAW-YOLOv11 offers an effective and lightweight solution for accurate bolt defect detection in real-world power transmission scenarios.