融合SAR分支深度学习的隧道病害识别与成像  

Identification and Imaging of Tunnel Lining Defects using Deep Learning Method with SAR Branch

作　　者：吴游宇[1] 刘德强 余飞[1] 徐乔 雷鸣 李博[2] WU You-yu;LIU De-qiang;YU Fei;XU Qiao;LEI Ming;LI Bo(CCCC Second Highway Consultants Co.Ltd.,Wuhan 430056,China;College of Control Science and Engineering,Shandong University,Jinan 250061,China)

机构地区：[1]中交第二公路勘察设计研究院有限公司,武汉市430056 [2]山东大学控制科学与工程学院,济南市250061

出　　处：《公路》2023年第12期320-327,共8页Highway

基　　金：湖北省重点研发计划项目,项目编号2021BAA185;武汉市重点研发计划项目,项目编号2022012202015071;中交二公院自主科研项目,项目编号KJFZ-2020-012。

摘　　要：随着交通事业的不断发展,大量隧道相继建设并陆续投入运营。在其运营过程中,隧道衬砌混凝土结构内部往往出现不可见的隐蔽病害,对工程安全带来了严重隐患。及时识别检测内部病害,预防安全事故的发生十分必要。探地雷达(Ground Penetrating Radar,GPR)是检测混凝土内部病害的主流无损检测技术,但探地雷达数据中的病害响应信号与实际结构内部病害形态并不存在直观的空间对应关系,从探地雷达数据中仅能估计病害的类型与大概位置,难以对其轮廓进行成像。针对上述问题,研究了基于探地雷达的隧道衬砌隐蔽病害智能识别技术,针对探地雷达数据特点,设计了融入合成孔径雷达(Synthetic Aperture Radar,SAR)成像处理分支的地下工程目标识别深度神经网络模型,通过在传统Unet网络中融入SAR成像分支,实现了对病害的准确成像,并成功对6种隧道衬砌常见的病害进行类型识别。采用仿真数据对该方法验证,结果表明本研究设计的融入SAR成像分支Unet网络的预测结果对6种病害的平均识别精度为96%,与传统Unet网络相比,精度提升了5%,由此证明本文构建的融入SAR成像分支的Unet网络有效提高了对病害的识别能力,能较好实现基于探地雷达数据的隧道衬砌隐蔽病害轮廓分割与类别识别。With the continuous development of the transportation industry,a large number of tunnels have been constructed and put into operation.In the course of its operation,invisible hidden defects often appear inside the lining concrete structures,which brings serious potential dangers to project safety.It is necessary to identify and detect internal defects in time to prevent the occurrence of accidents.Ground Penetrating Radar(GPR)is the main nondestructive testing technique for detecting internal defects of concrete,however,there is no direct spatial correspondence between the defect response signal in GPR data and the defect shape in the actual structure.Only the type and approximate position of the anomaly can be estimated from GPR data,it’s difficult to image its contours.In view of the above problems,in this paper the intelligent recognition technology of hidden defects in tunnel lining based on GPR is studied.Aiming at the characteristics of GPR data,a depth neural network model for underground engineering target recognition is designed,which is integrated with the branch of Synthetic Aperture Radar(SAR)imaging processing,the accurate imaging of abnormal objects is realized,and six types of abnormal objects are recognized successfully.The method is verified by simulation data.The results show that the average accuracy of the proposed method is 96%for six kinds of abnormal objects,compared with the conventional Unet network,the accuracy is improved by 5%,which proves that the Unet network integrated with the SAR Imaging Branch effectively improves the recognition ability of the anomaly body,it can realize the contour segmentation and classification of underground anomaly based on GPR data.

关 键 词：探地雷达 无损检测 内部病害识别 深度学习 合成孔径 

分 类 号：U456.3[建筑科学—桥梁与隧道工程]