摘要
随着我国隧道交通事业的快速发展,隧道日常养护的工作量急剧增加.利用探地雷达检测隧道衬砌病害因其经济、高效且无损等特点逐渐成为主流的方法,但从采集的海量雷达数据中准确的解释与识别病害特征更多的依赖人的经验,具有较强的主观性.为了提高衬砌病害的检测效率与准确率,本文提出了一种基于LSTM神经网络的隧道衬砌病害识别方法.该方法综合考虑了LSTM神经网络在图像识别领域的优势及隧道中探地雷达反射信号的特征,首次将LSTM神经网络应用在隧道衬砌病害检测与识别中.首先,利用FDTD数值模拟获得不同隧道病害模型对应的雷达波反射信号,分析出不同病害对应的反射特征.其次,通过改变病害的类型、厚度、埋深以及增加噪声等手段构建数据集,并进行随机拆分构建训练集、测试集与验证集.再次,基于LSTM搭建神经网络并优选出网络模型参数与网络结构,通过对比发现3层Bi-LSTM网络模型具有较高的准确率和较稳定的训练过程.最后,通过验证集数据对神经网络进行验证,并利用准确率、精确率及召回率三个指标评价验证结果.研究结果表明,基于LSTM神经网络的隧道衬砌病害识别方法能够快速、准确地识别渗水与空洞两种病害.本研究为今后隧道衬砌病害智能化识别方法的普及与发展提供了新的思路,也为LSTM神经网络的方法拓宽了应用领域.
With the rapid development of tunnel transportation in China,the workload of routine maintenance of tunnels has increased dramatically.The use of Ground Penetrating Radar(GPR)to detect tunnel lining diseases is becoming a common method due to its economic,efficient and non-destructive characteristics.But the reflection feature recognition and interpretation from massive radar data collected is time consuming and highly dependent on human experiences.In order to improve the detection efficiency and accuracy,this paper proposed an identification method of the tunnel lining diseases based on LSTM neural network.This method integrated the advantages of LSTM neural networks in the field of image recognition and the characteristics of GPR reflection signals from tunnel lining diseases.Then LSTM neural networks were applied to detect and recognize tunnel lining diseases for the first time.First,the reflected radar signals corresponding to different tunnel disease models were obtained by FDTD numerical simulation.Secondly,the dataset was constructed through different ways,such as changing the type,the thickness and the buried depth of the disease area,and adding noise to the radar signals.Then the whole dataset was randomly divided into the training dataset,the testing dataset and the validation dataset.Thirdly,a neural network was built based on LSTM and the network model parameters and structures were optimized.A conclusion was drawn that the three-layer Bi-LSTM network has higher accuracy and more stable training process.Finally,the neural network was verified,and three parameters(the accuracy rate,the precision rate and the recall rate)were used for the result evaluation.The results show that the identification method based on LSTM neural network can quickly identify two tunnel diseases with high accuracy and stable process.This research provides a new option for the popularization and development of intelligent identification methods of tunnel lining diseases in the future.It also broadens the application area of LSTM neural network.
作者
谢豪
王靖宇
XIE Hao;WANG JingYu(Beijing Key Laboratory of Earth Prospecting and Information Technology,China University of Petroleum(Beijing),Beijing 102249,China;National Key Laboratory of Petroleum Resources and Engineering,China University of Petroleum(Beijing),Beijing 102249,China)
出处
《地球物理学进展》
北大核心
2025年第5期2227-2236,共10页
Progress in Geophysics
基金
国家自然科学基金(41804143)
中国石油大学(北京)科研基金资助(2462016YJRC010,2462020YXZZ007)
油气资源与探测国家重点实验室自主研究课题(PRP/indep-4-1520)联合资助。
关键词
深度学习
LSTM神经网络
探地雷达
有限时域差分
智能识别
Deep learning
LSTM neural network
Ground penetrating radar
Finite difference time domain
Intelligent identification
