ArticlePublisher preview available

Intelligent recognition of ground penetrating radar images in urban road detection: a deep learning approach

Authors:
Fujun Niu at Chinese Academy of Sciences
Fujun Niu
Yunhui Huang
Yunhui Huang
Yunhui Huang
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Peifeng He
Peifeng He
Peifeng He
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Wenji Su
Wenji Su
Wenji Su
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Show all 6 authorsHide
Read publisher preview
Download citation
To read the full-text of this research, you can request a copy directly from the authors.

Abstract and Figures

In recent years, urban road collapse incidents have been occurring with increasing frequency, particularly in populous cities. To mitigate road collapses, geophysical prospecting plays a crucial role in urban road inspections. Ground Penetrating Radar (GPR), a non-destructive technology, is extensively employed for detecting urban road damage, with manual interpretation of GPR images typically used to identify buried objects. Nonetheless, manual interpretation methods are not only inefficient but also subjective, as they rely on the interpreter's experience, thereby affecting the interpreting reliability. This study investigates the distribution and causes of road collapses and develops a deep learning-based intelligent recognition model using GPR detection images of urban roads in cities of the South China as original samples. The finding reveal that road collapses are concentrated in the months of July and August, mainly caused by pipe leakage and rainfall. Common anomalies in urban road GPR detection comprise seven types of target objects, including cavity, pipeline, etc., with standard GPR images acquired through outdoor field experiments. Utilizing GPR forward simulation and image augmentation methods to expand the sample size, as well as generating anchor box dimensions through clustering analysis, have all been proven to effectively improve the model's performance. The urban road GPR image intelligent recognition model, based on the YOLOv4 algorithm, achieves a detection accuracy of up to 85%, proving effective in GPR detection of urban roads in cities of North China. This research offers valuable insights for the future application of deep learning-based image recognition algorithms in urban road GPR detection.
Figures - available from: Journal of Civil Structural Health Monitoring
This content is subject to copyright. Terms and conditions apply.
A preview of this full-text is provided by Springer Nature.
Learn more
Content available from Journal of Civil Structural Health Monitoring
This content is subject to copyright. Terms and conditions apply.
Vol.:(0123456789)
Journal of Civil Structural Health Monitoring (2024) 14:1917–1933
https://doi.org/10.1007/s13349-024-00818-5
123
ORIGINAL PAPER
Intelligent recognition ofground penetrating radar images inurban
road detection: adeep learning approach
FujunNiu1,2· YunhuiHuang1,2· PeifengHe1,2· WenjiSu1,2· ChenglongJiao1,2· LuRen1,2
Received: 31 July 2023 / Accepted: 31 May 2024 / Published online: 1 July 2024
© Springer-Verlag GmbH Germany, part of Springer Nature 2024
Abstract
In recent years, urban road collapse incidents have been occurring with increasing frequency, particularly in populous cities.
To mitigate road collapses, geophysical prospecting plays a crucial role in urban road inspections. Ground Penetrating Radar
(GPR), a non-destructive technology, is extensively employed for detecting urban road damage, with manual interpretation
of GPR images typically used to identify buried objects. Nonetheless, manual interpretation methods are not only ineffi-
cient but also subjective, as they rely on the interpreter's experience, thereby affecting the interpreting reliability. This study
investigates the distribution and causes of road collapses and develops a deep learning-based intelligent recognition model
using GPR detection images of urban roads in cities of the South China as original samples. The finding reveal that road
collapses are concentrated in the months of July and August, mainly caused by pipe leakage and rainfall. Common anoma-
lies in urban road GPR detection comprise seven types of target objects, including cavity, pipeline, etc., with standard GPR
images acquired through outdoor field experiments. Utilizing GPR forward simulation and image augmentation methods
to expand the sample size, as well as generating anchor box dimensions through clustering analysis, have all been proven
to effectively improve the model's performance. The urban road GPR image intelligent recognition model, based on the
YOLOv4 algorithm, achieves a detection accuracy of up to 85%, proving effective in GPR detection of urban roads in cities
of North China. This research offers valuable insights for the future application of deep learning-based image recognition
algorithms in urban road GPR detection.
Keywords Urban road collapse· Ground penetrating radar· Deep learning· Intelligent recognition· Sample expansion·
Clustering analysis
1 Introduction
In recent years, urban road collapse incidents have occurred
frequently, ranging from the arid and semi-arid regions in
northern China to the humid regions in southern China.
According to incomplete statistics, from 2011 to 2022, a
total of 885 urban road collapses occurred in China, with
an increasing trend year by year. The highest frequencies
were observed in populous cities such as Beijing, Shanghai,
and Zhengzhou (Fig.1). In densely populated cities, road
collapses can easily cause casualties and property losses,
thus attracting widespread attention from local government
departments and residents.
The direct cause of urban road collapse is the formation
and development of underground voids or cavities, lead-
ing to insufficient road bearing capacity [2]. The causes of
underground cavity formation can be divided into anthro-
pogenic and natural factors. Anthropogenic factors include
underground construction, groundwater extraction, and
pipeline leakage; natural factors include rainfall and geo-
logical conditions such as collapsible loess, karst geology,
and ancient river channels [1]. Urban road collapses exhibit
sudden characteristics [3]; however, the formation of under-
ground cavities is mostly not sudden but a slow development
process. Therefore, using geophysical methods to identify
the potential locations of underground cavities and other
structures in advance can effectively predict road sections
prone to collapse and prevent collapses from occurring.
* Fujun Niu
niu@scut.edu.cn
1 State Key Laboratory ofSubtropical Building Science, South
China University ofTechnology, Guangzhou510640, China
2 South China Institution ofGeotechnical Engineering, School
ofCivil Engineering andTransportation, South China
University ofTechnology, Guangzhou510640, China
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
... While numerical simulation is cost-effective, simulated images show characteristic differences from real images [50]. On-site collection, though costly, results in smaller datasets typically not exceeding 1000 images [51][52][53][54][55]. Therefore, compared to existing research, our dataset provides a larger-scale data foundation for internal road void image enhancement and intelligent recognition. ...
Intelligent Recognition of Road Internal Void Using Ground-Penetrating Radar
Article
Full-text available
  • Dec 2024
Internal road voids can lead to decreased load-bearing capacity, which may result in sudden road collapse, posing threats to traffic safety. Three-dimensional ground-penetrating radar (3D GPR) detects internal road structures by transmitting high-frequency electromagnetic waves into the ground and receiving reflected waves. However, due to noise interference during detection, accurately identifying void areas based on GPR-collected images remains a significant challenge. Therefore, in order to more accurately detect and identify the void areas inside the road, this study proposes an intelligent recognition method for internal road voids based on 3D GPR. First, extensive data on internal road voids was collected using 3D GPR, and the GPR echo characteristics of void areas were analyzed. To address the issue of poor image quality in GPR images, a GPR image enhancement model integrating multi-frequency information was proposed by combining the Unet model, Multi-Head Cross Attention mechanism, and diffusion model. Finally, the intelligent recognition model and enhanced GPR images were used to achieve intelligent and accurate recognition of internal road voids, followed by engineering validation. The research results demonstrate that the proposed road internal void image enhancement model achieves significant improvements in both visual effects and quantitative evaluation metrics, while providing more effective void features for intelligent recognition models. This study offers technical support for precise decision making in road maintenance and ensuring safe road operations.
View
... Non-destructive testing (NDE) techniques are essential in identifying structural sub-surface defects that are often missed during conventional visual inspections [18]. One such technology is ground-penetrating radar (GPR), which has been extensively utilized to detect material properties, corrosion-induced delamination and steel reinforcement corrosion [19][20][21][22]. Morries et al. employed ML models to characterize the relationships between GPR attributes and mechanical properties of concrete with various mix proportions [23]. ...
Ground penetrating radar-based automated defect identification of bridge decks: a hybrid approach
Article
Full-text available
  • Dec 2024
Nowadays, bridges play a crucial role, especially with the significant increase in the number of vehicles being driven worldwide. Hence, it is crucial to safeguard these structures from damage. This study aims to achieve this objective by proposing a novel hybrid framework for automated delamination detection of bridge decks based on ground penetrating radar (GPR), a mature technique utilized to localize underground deterioration or damage of bridges. The proposed framework comprises synchrosqueezed wavelet transform (SSWT), convolutional neural network (CNN), transfer learning, and metaheuristic optimization. First, original 1-D GPR signals undergo processing by SSWT to extract time–frequency characteristics that are sensitive to delamination. Next, extracted features are fed into deep CNN model VGG16 to develop a predictive model based on transfer learning. To enhance the generalization capability of the proposed model, modified whale optimization algorithm (MWOA) is utilized to optimize network hyperparameters during the training process. The performance of the proposed hybrid framework for delamination identification is validated using test data collected from the field testing of real bridges using GPR device. The proposed method demonstrates satisfactory results compared to other commonly used techniques, with the prediction accuracy of over 94%, providing an effective and efficient solution to the challenges of bridge defect detection.
View
Apple Detection in Complex Scene Using the Improved YOLOv4 Model
Article
Full-text available
  • Mar 2021
To enable the apple picking robot to quickly and accurately detect apples under the complex background in orchards, we propose an improved You Only Look Once version 4 (YOLOv4) model and data augmentation methods. Firstly, the crawler technology is utilized to collect pertinent apple images from the Internet for labeling. For the problem of insufficient image data caused by the random occlusion between leaves, in addition to traditional data augmentation techniques, a leaf illustration data augmentation method is proposed in this paper to accomplish data augmentation. Secondly, due to the large size and calculation of the YOLOv4 model, the backbone network Cross Stage Partial Darknet53 (CSPDarknet53) of the YOLOv4 model is replaced by EfficientNet, and convolution layer (Conv2D) is added to the three outputs to further adjust and extract the features, which make the model lighter and reduce the computational complexity. Finally, the apple detection experiment is performed on 2670 expanded samples. The test results show that the EfficientNet-B0-YOLOv4 model proposed in this paper has better detection performance than YOLOv3, YOLOv4, and Faster R-CNN with ResNet, which are state-of-the-art apple detection model. The average values of Recall, Precision, and F1 reach 97.43%, 95.52%, and 96.54% respectively, the average detection time per frame of the model is 0.338 s, which proves that the proposed method can be well applied in the vision system of picking robots in the apple industry.
View
Investigation on the phenomena and influence factors of urban ground collapse in China
Article
Full-text available
Urban ground collapse (UGC) is becoming more common in China, resulting in significant socioeconomic losses and even personal casualties. The frequency of UGC accidents is highest in the east coastal area owing to developed urbanization, while it is lowest in the northeast area because of its smallest land area. Natural causes (such as geological conditions and rainfall) and artificial causes (such as groundwater withdrawal, underground pipeline breakage, underground engineering, and other reasons) all contribute to UGC accidents in China. Groundwater influences most factors that lead to UGC. Adverse geology, such as collapsible loess and karst geology, is sensitive to groundwater. The groundwater environment is vulnerable to rainfall, pipeline leakage or groundwater withdrawal. Under the action of groundwater, the steady state of the soil may change, which finally leads to UGC. Groundwater control, which is essential for mitigating the risk of UGC, can be implemented through detailed geological surveys, sponge city and utility tunnel construction, and groundwater–level control measurement.
View
Yolo-v4 Deep Learning Model for Medical Face Mask Detection
Conference Paper
Full-text available
  • Mar 2021
The flora and fauna is facing a social disaster owing to the fast transfer of (Corona Virus). The disease with COVID-19 is mainly transmitted by respiratory droplets that are inhaled when people smell, talk, hack or sting. Wearing a veil is an incredible, powerful and easy way to prevent 82% of all respiratory infections. In this way, a number of cover protection structures and recognition structures have been put in place to provide compulsory provision for emergency clinics, air terminals, courier transport, sports offices, and retail outlets. Upper and lower tests have shown unusual strengths in specific real-world projects. Some of this may have been highlighted in the article. Ongoing revelations of articles relying on higher and lower reading models have yielded promising results by finding something found in the pictures. This paper focuses on a response to help to support a larger social request and to wear public covers using the revelation of the YOLO c4 object continuously engraved with images. The proposed Yolo v4 learning model has been pre-programmed with good program limitations. The organization ensures fast access that can deliver consistent results without resolving accuracy, or in complex arrangements. The proposed strategy will be divided into three categories: No dress, cover, and no veil. The model has outdone some of the proposed strategies in the past by gaining 99.98% accuracy during the preparation / testing.
View
Object Recognition for Organizing the Movement of Self-Driving Car
Article
Full-text available
  • Apr 2021
Today there is a revolution in the automotive industry. Cars are becoming self-driving with advanced sensors, cameras, and recognition algorithms. Algorithms and scenarios are evolving and improving every day, but it is too early for self-driving cars to go out on public roads. The work of recognition algorithms is far from ideal. For the correct and synchronized operation of all elements of an unmanned vehicle, a person needs to transfer all his intellectual experience to the computer systems of the vehicle. The developers are working to ensure that the car can see and understand what is happening around it. Such cars are already driving on the roads in test mode with a pilot, receiving a huge amount of information and learning. Already, tests and trials have proven that unmanned vehicles are safer than vehicles driven by people, and after mass implementation, the death rate on the roads will be reduced several times. Optical vision is an essential component of self-driving car due to the absence of human control. Speed, accurate detection of vehicles, empty parking, pedestrians, traffic signals, streets, and road signs can help self-driving vehicles drive safely and avoid mistakes and road accidents. On the other hand, object identification was challenging because objects in the physical world are affected by strong and low luminance, angularity, and scaling in image capture. The Convolutional Neural Network (CNN) approach was used by several researchers to improve object recognition outcomes in a variety of situations. However, the main disadvantage of this method is being unable to react quickly in real-time scenarios. In this work, the focus was on the YOLO model and the features of its work. Tests and comparisons with other state-of-the-art were given, which show that YOLO is the fastest model for real time objects recognition today. The architecture of the algorithm and the essential components on which the field experiment was conducted are described. Based on the results obtained, it was concluded that the YOLOv4 recognition model shows excellent results in terms of accuracy and speed.
View
Improved YOLOv4 Marine Target Detection Combined with CBAM
Article
Full-text available
  • Apr 2021
Marine target detection technology plays an important role in sea surface monitoring, sea area management, ship collision avoidance, and other fields. Traditional marine target detection algorithms cannot meet the requirements of accuracy and speed. This article uses the advantages of deep learning in big data feature learning to propose the YOLOv4 marine target detection method fused with a convolutional attention module. Marine target detection datasets were collected and produced and marine targets were divided into ten categories, including speedboat, warship, passenger ship, cargo ship, sailboat, tugboat, and kayak. Aiming at the problem of insufficient detection accuracy of YOLOv4’s self-built marine target dataset, a convolutional attention module is added to the YOLOv4 network to increase the weight of useful features while suppressing the weight of invalid features to improve detection accuracy. The experimental results show that the improved YOLOv4 has higher detection accuracy than the original YOLOv4, and has better detection results for small targets, multiple targets, and overlapping targets. The detection speed meets the real-time requirements, verifying the effectiveness of the improved algorithm.
View
Overview of seismic methods for Urban Underground Space
Article
Full-text available
  • Oct 2020
The development of Urban Underground Space (UUS) has become an effective way to solve the problem of urban space. Seismic methods have achieved good results in underground engineering construction, active faults detection, urban geological structure detection and other fields. Active source seismic methods in UUS include reflection methods, refraction methods, surface wave methods and vertical seismic profile (VSP) methods and so on. Ambient noise method is also rising gradually, which becomes an important method of urban underground space detection in the future, because it overcomes the problem that the active source method is interfered by urban noise. We briefly summarize these methods and introduce the application examples of UUS.
View
View
Attention-based YOLOv4 Algorithm in Non-destructive Radiographic Testing for Civic Aviation Maintenance
Preprint
  • Apr 2021
The aim of this paper is to use deep learning tools to innovate pre-trained object detection models to improve the accuracy of non-destructive testing (NDT) of civil aviation maintenance. First, this thesis classifies object defects for NDT, such as cracks, undercut, etc. Nowadays, thesis surveys innovation deep-learning methods technology is used to improve the defect detection performance inferencing capability, increase the accuracy and efficiency of automatic identification which in enhanced the safety and reliability of aircraft fuselage in future, mark hidden cracks and solve the challenges that cannot be identified by manual inspection. Second, recent mainstream techniques the YOLOv4 neural network to the graphics card GPU core operator to speed up the recognition of defect images is being applied to the non-destructive inspection process of aircraft maintenance on A, C and D-Level, fully validating the deep learning model's powerful defect detection target capability. The attention-based YOLOv4 algorithm is improved by applying a one-stage attention mechanism to the YOLOv4, thereby improving the accuracy of the innovation model. Finally, thesis improved YOLOv4 based on an attention mechanism is proposed for object detection NDT via the deep learning method to effectively improve and shorten the inspection anomaly detection method for automatic detection sensor systems.
View
GPR pattern recognition of shallow subsurface air voids
Article
  • May 2020
  • TUNN UNDERGR SP TECH
Countless subsurface voids in urban areas of cities threaten people’s lives and property. A workflow for automatically identifying subsurface voids from ground penetrating radar (GPR) data was developed in this study. The workflow consists of 3 stages: locating voids automatically from C-scans, then verifying voids from corresponding B-scans, and finally making judgements based upon the previous 2 sets of results. This study adopted 2 (Lai et al., 2016) approaches: approach 1 quantified the GPR response of air voids using forward modelling, while approach 2 used workflow prototyping and validation with inverse modelling. Forward simulations indicated that different ratios of void size to GPR signal footprint could result in a variety of patterns in B-scans: they can be hyperbolas, cross patterns, bowl shaped patterns and reverberations. With a database of void patterns of both C-scans and B-scans established, in approach 2 the workflow uses a pyramid pattern recognition method – with pixel value or gradient being used for feature identification – to search automatically for air-filled void responses in GPR data. The workflow was tested using 2 laboratory and field experiments and the results were promising. The constraint values proposed by the 2 experiments were validated with another site experiment. Given the huge workload involved in city-scale subsurface health inspections, a standardized workflow can help improve efficiency and effectiveness of subsurface void identification.
View
Semi-automatic detection of buried rebar in GPR data using a genetic algorithm
Article
  • Jun 2020
  • AUTOMAT CONSTR
This paper proposes a new method for semi-automatic detection of buried rebar in ground penetrating radar (GPR) data based on searching the reflected waveform of target by the genetic algorithm, which aims to overcome the inefficiency and instability of the manual process. Before the detection step, a binary image is generated by collecting all the local maximum points from A-scan GPR data to turn the searching for waveform image into curves. Meanwhile, the possible locations of the targets are preliminarily selected from the gray-scale frequency-wavenumber domain migration image of GPR data by setting a gray gradient threshold in the vertical direction. Then based on the genetic algorithm and time-distance analysis of the reflected signal, semi-automatic detection could be achieved by searching the best fitting hyperbola at every potential location in the binary image, and matching them with the reflected signal of targets according to several carefully chosen matching criteria. The impact of waveform distortion is also considered in the matching process to ensure a good adaptability of this method to the GPR data. Precision and efficiency of this method are analyzed lastly by four measured data of different antenna frequencies and targets, the average misjudgment and the undetected rate are 9.08% and 11.98% respectively. Meanwhile, the same GPR data is detected by four antenna central frequencies to find that the missed rate of semi-automatic detection decreases with the increase of the antenna frequency, when the misjudgment rate remains stable.
View