基于图像处理的桥梁橡胶支座剪切病害检测方法  

Image processing-based method for detecting shear damage in bridge rubber bearings

作　　者：梁栋 张少杰 周印霄 王莲香 张强 刘跃飞 LIANG Dong;ZHANG Shaojie;ZHOU Yinxiao;WANG Lianxiang;ZHANG Qiang;LIU Yuefei(School of Civil and Transportation Engineering,Hebei University of Technology,Tianjin 300401,China;Xingtai Road and Bridge Construction Co.,Ltd.,Xingtai Hebei 054000,China;Tianjin Highway Development and Service Center,Tianjin 300170,China;Tianjin Highway Engineering Design and Research Institute Co.,Ltd.,Tianjin 300171,China)

机构地区：[1]河北工业大学土木与交通学院,天津300401 [2]邢台路桥建设集团有限公司,河北邢台054000 [3]天津市公路事业发展服务中心,天津300170 [4]天津公路工程设计研究院有限公司,天津300171

出　　处：《北京交通大学学报》2023年第5期16-24,共9页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：国家自然科学基金(51978236);天津市交通运输委员会科技发展项目计划(2023-50)。

摘　　要：针对梁桥板式橡胶支座环境复杂、位置深入、难以手动测量剪切变形角度大小的问题,提出了一种基于图像自动计算橡胶支座剪切角度,并评估其剪切病害程度的方法.首先运用引入深度可分离卷积和多尺度注意力模块的U-Net网络完成图像中支座的识别和分割;其次对分割得到的支座二值图像使用简化的Alpha Shapes算法提取支座轮廓线;然后对支座轮廓进行凸包检测后提取凸包点及其坐标;最后对凸包点利用最小二乘法完成直线拟合,通过计算直线间夹角得到支座剪切角度并评估其剪切病害程度.研究结果表明:改进的U-Net模型支座分割的F1值和交并比均达到了95%以上;在天津市某桥梁检测中,利用本文方法对相机采集支座图像角度计算结果和人工实测结果进行对比,两者角度最大误差1.3°,检测的剪切病害等级分类结果相同.本文方法在橡胶支座剪切病害检测方面可实现非接触、自动化检测,为实际工程应用提供了参考.Addressing the complexities presented by the beam and bridge plate rubber bearings,in-cluding their challenging accessibility because of deep location and the intricate measurement of shear deformation angles,this study introduces an automated method for shear angle calculation and shear damage assessment based on image analysis.Firstly,a U-Net network integrated with depth-wise separable convolution and an Multi-scale Attention Module(MAM)is employed to identify and segment the bearings within the images.Secondly,the binary image of the segmented bearings is used for extracting bearing contour lines through a simplified Alpha Shapes algorithm.Then the con-vex packet detection is performed to extract the convex packet points and their respective coordi-nates.Finally,the least-squares method is utilized to fit the convex packet points into straight lines,quantifying the degree of shear damage through the calculation of the shear angle between these lines.The research results show that the improved U-Net model for bearing segmentation demon-strates F1 scores and Intersection over Union(IoU)exceeding 95%.In a bridge inspection con-ducted in Tianjin,this papers method is utilized to compare angle calculations from camera-captured bearing images with manual measurements.The maximum error between the two is re-corded at a mere 1.3°,and shear damage level classification produces consistent results.This papers method paves the way for non-contact,automatic shear damage detection in rubber bearings,pro-viding valuable insights for practical engineering applications.

关 键 词：桥梁工程 板式橡胶支座 剪切病害 图像处理 U-Net 

分 类 号：U445.71[建筑科学—桥梁与隧道工程]