Li Ruiping, Zhao Linchang, Wei Hao, Hu Guoqing, Xu Yongchi, Ouyang Bocheng, Tan Jin
School of Computer Science, Guiyang University, Guiyang, 550005, China.
Guizhou Provincial Key Laboratory for Digital Protection, Development and Utilization of Cultural Heritage, Guiyang, 550005, China.
Sci Data. 2025 Jul 1;12(1):1101. doi: 10.1038/s41597-025-05395-w.
This paper proposes the GYU-DET dataset for bridge surface defect detection, aiming to address the limitations of existing datasets in terms of scale, annotation accuracy, and environmental diversity. The GYU-DET dataset includes six types of defects: cracks, spalling, seepage, honeycomb surface, exposed rebar, and holes, with a total of 11,123 high-resolution images. It covers a variety of lighting and environmental conditions, comprehensively reflecting the diversity and complexity of bridge defects. The dataset provides comprehensive coverage of bridge structures, with images covering multiple key structural parts. Strict annotation guidelines ensure annotation accuracy and consistency, using the YOLO format, which facilitates model training and evaluation in computer vision tasks. To validate the effectiveness of the dataset, experiments were conducted using the YOLOv11 object detection model. The results show that GYU-DET can effectively support bridge defect detection tasks in the field of computer vision, providing high-quality data support for bridge surface defect detection tasks and promoting the development of intelligent bridge health monitoring technology.
本文提出了用于桥面缺陷检测的GYU-DET数据集，旨在解决现有数据集在规模、标注准确性和环境多样性方面的局限性。GYU-DET数据集包括六种类型的缺陷：裂缝、剥落、渗漏、蜂窝表面、露筋和孔洞，共有11123张高分辨率图像。它涵盖了各种光照和环境条件，全面反映了桥梁缺陷的多样性和复杂性。该数据集全面覆盖了桥梁结构，图像涵盖了多个关键结构部件。严格的标注指南确保了标注的准确性和一致性，采用YOLO格式，便于在计算机视觉任务中进行模型训练和评估。为了验证数据集的有效性，使用YOLOv11目标检测模型进行了实验。结果表明，GYU-DET能够有效地支持计算机视觉领域的桥梁缺陷检测任务，为桥面缺陷检测任务提供高质量的数据支持，推动智能桥梁健康监测技术的发展。
