LARNet-SAP-YOLOv11: A Joint Model for Image Restoration and Corrosion Defect Detection of Transmission Line Fittings Under Multiple Adverse Weather Conditions

Adverse weather conditions such as haze, rain, and snow degrade images captured by uncrewed aerial vehicles (UAVs) during transmission line inspections and severely affect the detection of corrosion defects in transmission line fittings. To address this challenge, we propose LARNet-SAP-YOLOv11, a unified end-to-end model that integrates lightweight image restoration and defect detection. The proposed model comprises LARNet, a lightweight all-in-one image restoration network, and SAP-YOLOv11, an enhanced object detector based on YOLOv11. LARNet is built upon the DehazeFormer architecture and introduces a Triplet Attention Block (TAB) to improve adaptability to various weather degradations. SAP-YOLOv11 enhances the baseline YOLOv11n by incorporating a Shallow Robust Feature Downsampling (SRFD) module, an Adaptive Fine-Grained Channel Attention (AFGCAttention) mechanism, and a Pixel-level Cross-Attention Feature Fusion (PCAFFusion) module, significantly improving corrosion area perception. Experimental results show that LARNet achieves an average PSNR of 30.43 dB and SSIM of 0.951 across different conditions. For defect detection, SAP-YOLOv11 improves the mAP@50 by 2.1% compared to the original YOLOv11n. When jointly applied, LARNet-SAP-YOLOv11 achieves an mAP@50 of 88.6%, outperforming the baseline YOLOv11n by 12.1% in challenging weather conditions. This unified model offers an efficient and reliable solution for UAV-based intelligent inspection of transmission lines under diverse environmental conditions.