【摘要】 表观裂缝检测是桥梁结构安全检测的重要内容，针对桥梁裂缝人工检测手段效率低以及基于一般的卷积神经网络（CNN）深度学习架构的裂缝检测模型耗时较大的问题，基于计算机视觉领域表现优异的深度学习架构Transformer，提出一种基于SegFormer语义分割网络的桥梁裂缝实时检测模型。研究表明，该模型是可行的，与基于CNN架构的LR-ASPP和BiSeNet V2等常用轻量级深度学习模型相比，裂缝检测的准确性、实时性与鲁棒性明显较优，将此模型结合无人机航拍应用于实际桥梁裂缝检测，取得了良好成效。更多还原

【Abstract】 The detection of surface cracks is a crucial aspect of bridge safety inspection. To address the issue of low efficiency of manual crack detection methods for bridges, and the time-consuming nature of conventional convolutional neural network(CNN)-based crack detection models, a real-time bridge crack detection model based on SegFormer semantic segmentation network is proposed. This model leverages the Transformer architecture, which has demonstrated excellent performance in computer vision.The study demonstrates the feasibility of the proposed model, which exhibits superior accuracy, real-time performance, and robustness in crack detection compared to commonly used lightweight CNN-based deep learning models such as LR-ASPP and BiSeNet V2. The proposed model, when combined with UAV aerial photography, has been applied to actual bridge crack detection, resulting in commendable outcomes.更多还原