Residual U-Net enhanced by dynamic receptive fields and attention mechanisms for crack detection in concrete protective structures of highway slopes

With the rapid expansion of highway networks in China, the stability of slopes in complex terrains such as mountainous and hilly regions has become a growing concern. Concrete protective structures are commonly used to safeguard highway slopes from geological hazards such as landslides and collapses. However, prolonged exposure to environmental factors and external loads makes these structures prone to cracking, which poses significant risks to highway safety. To address this issue, this paper proposes a residual U-Net model enhanced by dynamic receptive fields and attention mechanisms (RDAU-Net). Building upon the classic U-Net architecture, the model integrates deep residual convolution units, dynamic receptive fields (DRFs) and channel-spatial attention mechanisms to improve crack segmentation accuracy and robustness. After training and optimising on the publicly available DeepCrack dataset, the model surpasses five other state-of-the-art crack segmentation algorithms in key metrics, including precision (Pr), recall (Re), F 1 score (F 1 ) and intersection over union (IoU), demonstrating superior generalisation capability. Furthermore, when it is applied to crack detection in concrete protective structures of in-service highway slopes, the proposed method maintains high accuracy and stability. These findings suggest that the proposed method offers a promising approach for intelligent crack detection and safety monitoring in highway slope protective structures.