基于无人机的悬索桥主缆自动巡检及小样本表观病害识别  

Automatic Inspection of Main Cables of Suspension Bridge Based on UAV and Apparent Defect Identification with Small-sized Samples

作　　者：冯东明[1,2,3] 余星宇 黎剑安 吴刚 FENG Dong-ming;YU Xing-yu;LI Jian-an;WU Gang(Key Laboratory of Concrete and Prestressed Concrete Structures of the Ministry of Education,Southeast University,Nanjing 21l189,Jiangsu,China;National and Local Joint Engineering Research Center for Intelligent Construction and Maintenance,Southeast University Nanjing 21l189,Jiangsu,China;National Key Laboratory of Safety,Durability and Healthy Operation of Long Span Bridges,Nanjing 211189,Jiangsu,China;School of Civil Engineering,Southeast University,Nanjing,211189,Jiangsu,China)

机构地区：[1]东南大学混凝土与预应力混凝土结构教育部重点实验室,江苏南京211189 [2]东南大学智慧建造与运维国家地方联合工程研究中心,江苏南京2211189 [3]长大桥梁安全长寿与健康运维全国重点实验室,江苏南京211189 [4]东南大学土木工程学院,江苏南京211189

出　　处：《中国公路学报》2024年第2期29-39,共11页China Journal of Highway and Transport

基　　金：中央高校基本科研业务费专项资金项目(2242022k30030);江苏省交通运输科技项目(2021Y15);东南大学新进教师科研启动经费项目(RF1028623149)。

摘　　要：为了实现悬索桥主缆的自动化、智能化检查,开展了基于无人机的主缆巡检路径规划和小样本数据下的主缆病害识别研究。首先,利用无人机倾斜摄影测量技术快速建立悬索桥的三维模型,提出主缆无人机自动巡检路径的规划方法;然后,采用Faster RCNN网络模型识别主缆图像中的表观病害;最后,采用基于图像融合的数据增强方法,提高小样本数据集下目标检测的准确率。在Faster RCNN网络模型训练过程中,随着训练轮次的增加,测试集中裂纹、锈蚀和划痕3类病害的平均精确率得到提升,并在第15个训练轮次后逐渐稳定,在经过100个训练轮次后,测试集中所有类别的平均精确率为0.723。以小龙湾桥为研究对象,进行了主缆的现场检查试验。研究结果表明:基于悬索桥三维模型进行主缆无人机自动巡检路径规划具有实际可行性;基于Faster RCNN网络模型能较准确地识别主缆的裂纹、锈蚀和划痕病害;利用图像融合方法生成病害数据能有效克服数据样本少的问题,并提高识别的准确性。To realize automatic and intelligent inspection of main cables of a suspension bridge,a route planning method for cable inspection using an unmanned aerial vehicle(UAV)and apparent defect identification with small-sized samples is proposed.First,UAV oblique photogrammetry is utilized to rapidly construct a three-dimensional model of the targeted suspension bridge,facilitating a proposed automatic route planning for UAV inspection of main cables.Subsequently,the Faster R-CNN neural network model is employed to identify apparent defects such as cracks,corrosion,and scratches from images of the main cables.Finally,an image fusion-based data augmentation method is used to improve the accuracy of defect detection with a small-sized sample dataset.During the training process of the Faster R-CNN neural network model,the average accuracy of the three types of defects(i.e.,cracks,corrosion,and scratches)in the test dataset increases with the increase of the number of training epochs and gradually stabilizes after the 15th epoch.After 100 training epochs,the average accuracy for the three types of defects in the test dataset reaches 0.723.Field main cable inspections were conducted on the Xiaolongwan Bridge,and the results indicate that automatic UAV route planning for the inspection of main cables based on the established three-dimensional model is feasible in practice.The Faster R-CNN network model can accurately identify cracks,corrosion,and scratches in the main cables.The proposed fusion-based data augmentation method can effectively enhance the defect identification accuracy from small-sized samples.

关 键 词：桥梁工程 病害检测 深度学习 主缆检测 路径规划 

分 类 号：U446.2[建筑科学—桥梁与隧道工程]