Gradient transformer Self-Attention U-Net for enhanced crack detection in concrete bridges

Concrete bridge maintenance is crucial for infrastructure sustainability and public safety. Accurate identification of cracks in bridge components is a critical task, yet traditional inspection methods often fall short due to their limitations in accuracy and efficiency. This paper introduces the Gradient Transformer Attention U-Net (GTAU-Net) model, a novel deep learning approach that significantly advances crack detection in bridge components. Unlike conventional attention-based U-Nets, GTAU-Net introduces a Quantum Fused Filter (QFF) to pre-process images by integrating multiple edge and gradient patterns through a quantum-inspired hybrid filtering strategy. It further computes Gradient Saliency Scores (GSS) to dynamically guide the self-attention mechanism, enabling more precise localization and feature extraction. Through this dual enhancement, GTAU-Net effectively handles varying crack sizes, shapes, orientations, and environmental conditions. Experimental results demonstrate that GTAU-Net achieves an impressive 99.42% accuracy significantly outperforming existing models and measures crack lengths, highlighting sustainability. This research contributes to the advancement of automated crack detection technology, offering a promising solution for enhancing infrastructure safety and durability. To promote transparency and reproducibility, the code and dataset used in this study are publicly available at Zenodo: https://doi.org/10.5281/zenodo.15617661 .