基于U-Net和数学形态学的混凝土桥梁病害定量识别方法研究  

Research on Quantitative Identification Methods for Concrete Bridge Defects Based on U-Net and Mathematical Morphology

作　　者：黄彩萍 田旺源[1] 李青 HUANG Caiping;TIAN Wangyuan;LI Qing(School of Civil Engineering,Architecture and Environment,Hubei University of Technology,Wuhan 430068,China;Key Laboratory of Intelligent Health Perception and Ecological Restoration of Rivers and Lakes,Ministry of Education,Hubei University of Technology,Wuhan 430068,China;Huzhou Highway and Transportation Management Center,Huzhou 313000,China)

机构地区：[1]湖北工业大学土木建筑与环境学院,湖北武汉430068 [2]湖北工业大学河湖健康智慧感知与生态修复教育部重点实验室,湖北武汉430068 [3]湖州市公路与运输管理中心,浙江湖州313000

出　　处：《桥梁建设》2025年第1期64-71,共8页Bridge Construction

基　　金：国家自然科学基金项目(51708188)。

摘　　要：为使桥梁病害检测更加高效、客观和智能,提出一种自动识别并定量计算混凝土病害尺寸的方法。该方法采用视觉几何组网络(Visual Geometry Group Network,VGG)作为U形网络(U-Net)的主干网络,对混凝土病害(剥落、裂缝和露筋)图像进行语义分割,采用数学形态学算法对图像中的病害区域进行优化。通过MATLAB软件计算得到优化后的分割图像中病害区域像素点的数量,并利用参照物标定出图像中单个像素点的尺寸,计算得到混凝土病害的面积(或长度)。采用该方法对河南省许昌市17座现役钢筋混凝土桥梁病害图像进行语义分割实验。结果表明:U-Net能以较高的精度对复杂背景下混凝土桥梁多类病害进行像素级的分类,类别平均像素准确率为90.53%,平均交并比为80.54%。使用数学形态学对语义分割图像进行优化后,计算精度明显提高,优化后的误差绝对值为0.08%~0.21%。This paper presents a method that can automatically identify concrete defects and quantitatively calculate the sizes of concrete defects,aiming to allow the detection of bridge defects to be more efficient,objective and intelligent.In this method,Visual Geometry Group Network(VGG)is selected as the backbone network of U-Net to perform semantic segmentation on the concrete defects images(spall,crack and exposed rib),and mathematical morphology algorithm is utilized to optimize the defects region in the image.The number of pixels in the defects area in the optimized segmentation image was calculated by MATLAB software,and the size of a single pixel in the image was calibrated by reference objects,and the area(or length)of the concrete defects were calculated.The semantic segmentation of defect images from 17 in-service reinforced concrete bridges in Xuchang City,Henan Province demonstrates that U-Net can classify various types of concrete bridge defects under complex background with high accuracy at pixel level,the average pixel accuracy of each category is 90.53%,and the average crossover ratio is 80.54%.After using mathematical morphology to optimize the semantic segmentation image,the calculation accuracy is markedly improved,and the absolute error after optimization is between 0.08%and 0.21%.

关 键 词：混凝土桥梁 U-Net 数学形态学 语义分割 定量计算 病害识别 

分 类 号：U445.71[建筑科学—桥梁与隧道工程] TP183[交通运输工程—道路与铁道工程]