摘要
桥梁结构表面裂缝检测为桥梁状态识别、病害治理、安全评估提供了重要状态信息和决策依据。为解决传统人工检测方法存在的危险性高、影响交通、费用昂贵等问题,提出基于无人机(Unmanned Aerial Vehicle, UAV)及深度学习的桥梁结构裂缝智能识别方法。采用大疆M210-RTK多旋翼无人机进行贴近航摄,获取桥梁结构混凝土表面高清图像;利用SDNET裂缝数据集等图像资源,制作1 133张标记裂缝精确区域的深度学习训练样本图像库;引入掩膜区域卷积神经网络(Mask R-CNN)深度学习算法,训练和建立Mask R-CNN裂缝识别模型;基于Mask R-CNN裂缝识别模型,采用矩形滑动窗口模式扫描混凝土表面高清图像,实现裂缝自动识别和定位。构建包含图像二值化、连通域去噪、边缘检测、裂缝骨架化、裂缝宽度计算等流程的图像后处理方法,实现裂缝形态及宽度信息自动获取。通过精度验证试验,证实采用M210-RTK无人机+ZENMUSE X5S相机+45 mm奥林巴斯镜头的组合装备,当无人机至桥梁结构表面垂直距离为10.0 m时,无人机方法识别的裂缝宽度与裂缝测量仪结果吻合,其绝对误差小于0.097 mm,相对误差小于9.8%。将该无人机裂缝检测方法应用于高136.8 m长沙市洪山大桥桥塔表面裂缝检测,采用深度学习Mask R-CNN算法进行裂缝智能识别,其裂缝识别准确率和召回率分别达到92.5%和92.5%。研究结果表明:无人机桥梁裂缝检测方法可实现高耸桥梁结构表面裂缝的远程、非接触、自动化检测,具有重要的科学研究和工程应用价值。
Surface crack detection of bridge structures provides important condition information and decision basis for condition identification, disease regulation, and safety assessment of bridge structures. To address the problems of traditional manual detection methods, such as high risk, traffic impact and high cost, an intelligent crack identification method for bridge structures is proposed based on unmanned aerial vehicles(UAVs) and deep learning. A multi-rotor DJI M210-RTK UAV was used to acquire high-resolution images of the concrete surfaces of bridge structures. Using the SDNET crack dataset and other image resources, a training image library for the deep-learning was established for deep learning, covering 1133 images of precisely marked crack aeras. The deep-learning recognition model was trained and established using the mask region-based convolutional neural network(Mask R-CNN) algorithm. Based on the Mask R-CNN identification model, cracks were automatically identified and located by scanning high-resolution concrete surface images in rectangular sliding windows. A post-processing procedure was developed to identify crack shapes and widths, covering image binarization, connected domain denoising, edge detection, crack skeletonization and width calculation. A set of instruments consisting of a DJI M210-RTK UAV, a ZENMUSE X5 S camera and an Olympus lens with a focal length of 45 mm was used in the verification experiment. The crack widths identified by the UAV method agree well with those measured by the crack width test equipment when the distance from the UAV to the surface of the bridge structure was 10 m. The absolute errors are less than 0.097 mm, and the relative errors are less than 9.8%. The UAV method was applied to the surface crack detection of the bridge tower of the Changsha Hongshan Bridge, with a height of 136.8 m. The deep learning of the Mask R-CNN algorithm was used to identify cracks automatically from the images with an accuracy and a recall rate of 92.5% and 92.5%, respectively. The UAV method, which is highly efficient, safe, and inexpensive, is capable of remote, non-contact and automatic detection of cracks in high-rise bridge structures, which has important value for scientific research and engineering applications.
作者
余加勇
李锋
薛现凯
朱平
吴鑫赟
卢培升
YU Jia-yong;LI Feng;XUE Xian-kai;ZHU Ping;WU Xin-yun;LU Pei-sheng(Key Laboratory of Wind Engineering and Bridge Engineering in Hunan Province,Hunan University,Changsha 410082,Hunan,China;College of Civil Engineering,Hunan University,Changsha 410082,Hunan,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2021年第12期80-90,共11页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2016YFC0800207)
长沙市科技计划项目(kq1907110)。