摘要
路面病害检测一直是保障公路施工质量的一大挑战。探地雷达(GPR)作为一种无损检测(NDT)工具,现已被广泛应用于沥青路面的施工监测和评估,但传统的探地雷达图像依靠人工检测,效率较低。深度学习检测算法已证明其能以接近实时的速度从图像和视频中识别各种物体,而目前自动探地雷达图像检测的研究应用很少。为了解决人工识别病害图像的缺陷,提出一种基于Faster-RCNN的GPR病害图像检测模型。样本选取自中柬共建“一带一路”项目金港高速公路上的一个试验路段,利用安装在无人机上的探地雷达检测路面状况。模型利用实地采集的雷达图像进行训练和测试,并采用准确率、召回率和综合平均精度(mAP)评价分类和检测结果。测试结果表明,标记各类病害图像识别的准确率和召回率均>91%,mAP为94.1%。Faster-RCNN模型能准确和定量地识别出探地雷达检测中的裂缝、空洞和松散图像,满足高速公路施工质量检测需要。
Pavement distress detection has always been a major challenge in ensuring the quality of highway construction.Ground penetrating radar(GPR)is now widely used as a non-destructive testing(NDT)tool for construction monitoring and assessment of asphalt pavements,but traditional ground penetrating radar images rely on manual inspection,which is less efficient.Deep learning detection algorithms demonstrate their ability to identify a variety of objects from images and videos at near realtime speeds,while there are currently few research applications for automatic ground-penetrating radar image detection.In order to solve the defects of artificial recognition of disease images,a detection model of GPR disease images based on Faster-RCNN is proposed.The samples are taken from a test section of the PPSHV expressway,a China-Cambodia“Belt and Road”Initiative project,where ground-penetrating radar mounted on a drone is used to detect road conditions.The model is trained and tested with the acquired radar images in the field,using accuracy rate,recall rate and combined average precision(mAP)to evaluate the results of classification and detection.The test results show that the accuracy and recall rate of marking various types of disease image recognition are over 91%,with an mAP of 94.1%.The Faster-RCNN model can accurately and quantitatively detect cracks,voids and loose images in ground-penetrating radar inspection,meeting the needs of quality inspection in highway construction.
作者
房振华
谭治海
朱哲
张静晓
FANG Zhenhua;TAN Zhihai;ZHU Zhe;ZHANG Jingxiao(China Road&Bridge Corporation,Beijing 100011,China;School of Civil Engineering and Geomatics,Southwest Petroleum University,Chengdu,Sichuan 610500,China;School of Economics and Management,Chang’an University,Xi’an,Shaanxi 710068,China)
出处
《施工技术（中英文）》
CAS
2023年第24期76-82,共7页
Construction Technology
基金
中国路桥工程有限责任公司科研项目(2020-zlkj-04)。
关键词
道路工程
路面
图像检测
探地雷达
病害检测
区域卷积神经网络
roads
pavements
image detection
ground penetrating radar(GPR)
disease detection
regional convolutional neural network
