Abstract: Concrete cracks widely exist on the surfaces of bridges, roads, and buildings, affecting engineering safety. Therefore, crack damage detection of concrete structures is an important indicator for evaluating infrastructure safety. To address the problems such as low accuracy in traditional manual inspection, large parameter volumes and blurry crack details in existing crack detection models, a high-precision crack detection method for wall-climbing robots based on the YOLO-DSD network was proposed. The proposed method was based on the instance segmentation framework of the YOLO11-seg model, integrating the C3K-DWR module, the attention M-SEAM module and the Dyhead segmentation head. The YOLO-DSD algorithm reduced computational load while maintaining high-precision crack segmentation, and its accuracy surpassed that of mainstream segmentation models in the YOLO series. The research results show that compared with the baseline model, the precision of the proposed method (Precision) reaches 80.5%, with an improvement of 7.2%, and mAP50 reaches 71.6%, with an increase of 5.4%. Finally, the YOLO-DSD algorithm is deployed on a wall-climbing robot for inspection, enabling accurate extraction of crack edge information in complex backgrounds, which provides an efficient, robust and practical crack detection solution for real engineering applications.

Key words: bridge engineering; instance segmentation; concrete cracks; YOLO11-seg; wall-climbing robot

摘要： 混凝土裂缝广泛存在于桥梁、道路及建筑物表面，严重影响着工程安全，因此混凝土结构裂缝损伤检测是评价基础设施安全性的重要指标。针对传统人工检测精度低、现有裂缝检测模型参数量大且裂缝细节模糊等问题，提出了一种基于YOLO-DSD算法的爬壁机器人高精度裂缝检测方法。该方法基于YOLO11-seg模型的实例分割框架，融合了C3K-DWR、注意力M-SEAM和分割头Dyhead模块；YOLO-DSD算法在保持高精度分割裂缝的同时减少了计算量，且准确率超过了YOLO系列主流的分割模型。研究结果表明：相较于基准模型，其精确率Precision达到80.5%，提升了7.2%，mAP50达到71.6%，提升了5.4%。最后将YOLO-DSD算法部署于爬壁机器人进行巡检，在复杂背景下能精准提取裂缝边缘信息，为实际工程应用的裂缝检测提供了一种高效、鲁棒性且实用的裂缝检测方案。

关键词: 桥梁工程；实例分割；混凝土裂缝；YOLO11-seg；爬壁机器人