面向桥梁拉吊索护套的三阶段表观病害检测流程

刘玉静; 栗竞帆; 范毅; 张佰林; 朱尧于; 魏晓晨; 刘宇飞

doi:10.3724/j.gyjzG25062002

面向桥梁拉吊索护套的三阶段表观病害检测流程

doi: 10.3724/j.gyjzG25062002

1. 江苏高速公路工程养护技术有限公司, 南京 210049;
2. 清华大学土木工程系, 北京 100084;
3. 江苏润扬大桥发展有限责任公司, 南京 210001;
4. 江苏泰州大桥有限公司江苏泰州 225321;
5. 中交公路长大桥建设国家工程研究中心有限公司, 北京 100120

详细信息

作者简介:
刘玉静，硕士，高级工程师，主要从事桥隧养护技术研究。

通讯作者:
刘宇飞，博士，副研究员，主要从事工程结构性态感知识别与评价预测研究，liuyufei@tsinghua.edu.cn。

计量
- 文章访问数: 70
- HTML全文浏览量: 13
- PDF下载量: 8
- 被引次数: 0
出版历程
- 收稿日期: 2025-06-20
- 网络出版日期: 2025-09-12

A Three-Stage Surface-Defect Detection Pipeline for Bridge Stay-Cable Sheaths

1. Jiangsu Expressway Engineering Maintenance Technology Corporation, Nanjing 210049, China;
2. Department of Civil Engineering, Tsinghua University, Beijing 100084, China;
3. Jiangsu Runyang Bridge Development Corporation, Nanjing 210001, China;
4. Jiangsu Taizhou Bridge Corporation, Taizhou 225321, China;
5. CCCC National Engineering Research Center of Long Highway Bridge Construction Corporation, Beijing 100120, China

摘要

摘要: 拉吊索护套的表观病害直接关系桥梁结构的安全冗余与使用寿命。然而，大景深成像、复杂背景及索体长度带来的信息割裂，仍制约着拉吊索护套表观病害自动化检测精度。针对现有研究中数据预处理不足与整体性分析缺失的问题，构建了一套涵盖实例分割背景去除、图像拼接、YOLO检测验证三阶段智能检测流程。首先，设计了基于微调后的U²-Net对爬索机器人实采图像进行像素级裁剪，消除背景干扰并保留护套精细特征；随后，采用改进区域SIFT特征匹配实现纵向展开与横向拼接，获得索体图像拼接结果；最后通过针对护套表观病害任务改进后的YOLOv5目标检测模型验证上述流程的有效性。在涵盖1500幅原始缺陷图像的自建数据集上，修改后的模型mAP@0.5指标相比基础YOLOv5提升2.8%。三阶段检测流程能够有效提升拉吊索护套病害的检测精度，并为后续拉吊索病害的三维定位与寿命评估提供高质量数据支撑。
- 拉吊索护套 /
- 病害检测 /
- 图像拼接 /
- 实例分割 /
- 智能检测流程
Abstract: Surface defects of stay-cable sheaths directly jeopardize the structural redundancy and service life of cable-supported bridges. However， large depth-of-field imaging， cluttered backgrounds， and considerable cable length fragment visual information still impairs the accuracy of automated inspection accuracy. To remedy the inadequate data preprocessing and absence of holistic analysis in the existing work， an intelligent three-stage inspection pipeline that integrates instance-segmentation-based background removal， image mosaicking， and YOLO-based validation was developed. First， a fine-tuned U²-Net performed pixel-level cropping on images captured by a cable-climbing robot， suppressing background noise while preserving sheath details. Next， an enhanced regional SIFT matcher conducted longitudinal unfolding and transversed stitching to generate a seamless panoramic view of the cable surface. Finally， a task-oriented， improved YOLOv5 detector—augmented with lightweight attention and loss-function refinements—verified the effectiveness of the workflow for sheath-defect recognition. Experiments on a self-built data set of 1500 original images showed that the proposed model raised the mAP@0.5 by 2.8% comparing to the baseline YOLOv5 and maintained real-time inference. The three-stage pipeline markedly enhanced stay-cable sheath defect detection precision and supplied high-quality data for subsequent three-dimensional localization and service-life evaluation.
- stay-cable sheath /
- defect detection /
- image stitching /
- instance segmentation /
- intelligent inspection pipeline

HTML全文

参考文献(20)

[1]	KONG X，LIU Z，LIU H，et al. Recent advances on inspection，monitoring，and assessment of bridge cables［J］. Automation in Construction，2024，168，105767.
[2]	LIU Z，GUO T，HEBDON M H，et al. Corrosion fatigue analysis and reliability assessment of short suspenders in suspension and arch bridges［J］. Journal of Performance of Constructed Facilities，2018，32（5），04018060.
[3]	韩亮，樊健生. 近年国内桥梁垮塌事故分析及思考［J］. 公路，2013（3）:124-127.
[4]	BRODIE G H，CHAN J J，MICHAEL B. Towards UAV-based bridge inspection systems:a review and an application perspective［J］. Structural Monitoring and Maintenance 2015，2（5）:283-300.
[5]	HOU S，DONG B，WANG H，et al. Inspection of surface defects on stay cables using a robot and transfer learning［J］. Automation in Construction，2020，119，103382.
[6]	HUANG Z，GUO B，DENG X，et al. Trans-DCN:a high-efficiency and adaptive deep network for bridge cable surface defect segmentation［J］. Remote Sensing，2024，16（15），2711.
[7]	吴华勇，孟庆领，王枫，等. 基于深度学习算法的桥梁拉索护套表观病害识别［J］. 建筑科技，2024，8（8）:37-42.
[8]	DING N，ZHENG Z，SONG J，et al. CCRobot-III:a split-type wire-driven cable climbing robot for cable-stayed bridge inspection［C］// 2020 IEEE International Conference on Robotics and Automation（ICRA）. IEEE，2020:9308-9314.
[9]	本刊. 更轻快更智能更精准第六代桥梁拉索检测机器人首次应用于跨江桥梁［J］. 西部交通科技，2022（5）:209.
[10]	LI X，GAO C，GUO Y，et al. Cable surface damage detection in cable-stayed bridges using optical techniques and image mosaicking［J］. Optics & Laser Technology，2019，110:36-43.
[11]	QIN X，ZHANG Z，HUANG C，et al. U²-Net:Going deeper with nested U-structure for salient object detection［J］. Pattern Recognition，2020，106，107404.
[12]	LOWE D G. Object recognition from local scale-invariant features［C］// Proceedings of the seventh IEEE international conference on computer vision，IEEE，1999，2:1150-1157.
[13]	LOWE D G. Distinctive image features from scale-invariant keypoints［J］. International Journal of Computer Vision，2004，60:91-110.
[14]	REDMON J. You only look once:Unified，real-time object detection［C］// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016.
[15]	崔晓宁，王起才，李盛，等. 基于YOLO-v5的双块式轨枕裂缝智能识别［J］. 铁道学报，2022，44（4）:104-111.
[16]	施辉，陈先桥，杨英. 改进YOLO v3的安全帽佩戴检测方法［J］. 计算机工程与应用，2019，55（11）:213-220.
[17]	胡晓伟，闫奕昕，王大为，等. 基于YOLOM算法的路面病害轻量化检测方法［J/OL］. 中国公路学报，1-15［ 2024-11-19］. http://kns.cnki.net/kcms/detail/61.1313.U.20241108.1337.007.html.
[18]	CHO Y J. Weighted intersection over union（wIoU）:a new evaluation metric for image segmentation［J］. arXiv preprint arXiv:2107.09858，2021.
[19]	LI C，ZHOU A，YAO A. Omni-dimensional dynamic convolution［J］. arXiv preprint arXiv:2209.07947，2022.
[20]	OUYANG D，HE S，ZHANG G，et al. Efficient multi-scale attention module with cross-spatial learning［C］// ICASSP 2023-2023 IEEE International Conference on Acoustics，Speech and Signal Processing（ICASSP）. IEEE，2023:1-5.