基于因果学习的高铁接触网吊弦跨域鲁棒性检测研究OA

With the rapid development of China's high-speed railway infrastructure,the operational stability of the overhead contact system(OCS)has become increasingly critical.The structural integrity and operational status of catenary droppers,as key components of the OCS power supply system,directly affect train safety and operational reliability.To achieve automated and intelligent inspection of dropper health conditions,an essential task is to ensure accurate detection under complex environmental conditions.This paper proposed a cross-domain object detection algorithm,based on causal learning for efficient recognition of catenary droppers in challenging scenarios.First,multiple data augmentation strategies-including global transformation,geometric deformation,and environmental simulation were employed to expand the source-domain samples and simulate diverse real-world working conditions.These augmentations enabled the effective transfer of label information.Second,a causal attention module was integrated into an improved Faster R-CNN detection framework to explore the causal relationships between critical regions and global features.This strengthened the representation of dropper-related features while suppressing background noise and irrelevant information.Finally,a causal prototype learning method maps features of the same category in both source and target domains into a unified prototype space,thereby achieving cross-domain feature alignment and mitigating detection errors caused by source-domain bias and sample imbalance.The proposed method is extensively validated across diverse and representative environments.Experimental results show that it consistently outperforms mainstream cross-domain object detection methods,such as Faster R-CNN and YOLOv11,achieving superior accuracy on the test set.In addition to maintaining high detection precision,the method can demonstrate strong cross-domain adaptability and robustness,highlighting its promising potential for practical engineering applications.