Road Detection, Monitoring, and Maintenance Using Remotely Sensed Data

Abstract

Roads are a form of critical infrastructure, influencing economic growth, mobility, and public safety. However, the management, monitoring, and maintenance of road networks remain a challenge, particularly given limited budgets and the complexity of assessing widespread infrastructure. This Special Issue on “Road Detection, Monitoring, and Maintenance Using Remotely Sensed Data” presents innovative strategies leveraging remote sensing technologies, artificial intelligence (AI), and non-destructive testing (NDT) to optimize road infrastructure assessment. The ten papers published in this issue explore diverse methodologies, including novel deep learning algorithms for road inventory, novel methods for pavement crack detection, AI-enhanced ground-penetrating radar (GPR) imaging for subsurface assessment, high-resolution optical satellite imagery for unpaved road assessment, and aerial orthophotography for road mapping. Collectively, these studies demonstrate the transformative potential of remotely sensed data for improving the efficiency, accuracy, and scalability of road monitoring and maintenance processes. The findings highlight the importance of integrating multi-source remote sensing data with advanced AI-based techniques to develop cost-effective, automated, and scalable solutions for road authorities. As the first edition of this Special Issue, these contributions lay the groundwork for future advancements in remote sensing applications for road network management.

1. Introduction

Roads are fundamental to economic growth, social connectivity, and public safety, yet their management presents persistent challenges. The need for effective road monitoring and maintenance has never been greater, as increasing traffic volumes and climate-related deterioration put existing infrastructure at risk. Traditional inspection methods rely on labor-intensive, time-consuming processes that are both costly and prone to human error. To address these issues, the integration of remote sensing technologies, AI, and NDT methods has gained momentum, providing scalable and cost-effective solutions for infrastructure management.

This Special Issue, “Road Detection, Monitoring, and Maintenance Using Remotely Sensed Data”, brings together ten studies that demonstrate cutting-edge methodologies for road condition assessment, monitoring, and maintenance. These papers explore diverse approaches—including AI-driven crack detection, machine learning-based road condition prediction, subsurface imaging with (GPR), high-resolution optical satellite imagery, and deep learning applications for road mapping—to enhance the efficiency and accuracy of road infrastructure management.

In this introduction, we provide a comprehensive overview of how these ten contributions collectively advance the field. By structuring this discussion around four key thematic areas, namely road inventory and mapping, crack detection and pavement condition assessment, subsurface imaging, and emerging technologies in road monitoring, we concisely highlight the impact of each study and its broader implications for the future of infrastructure maintenance and safety.

In the List of Contributions section is a list of the articles published in this Special Issue.

1.1. Road Inventory and Mapping: The Shift Toward Automated, Large-Scale Approaches

One of the biggest challenges in road management is the ability to efficiently inventory, map, and monitor road networks, particularly in large or remote areas. Traditional road surveys are slow and expensive, often requiring manual data collection. This Special Issue presents multiple studies that leverage AI and remote sensing for scalable, automated road mapping.

Tardy et al. introduce a low-cost mobile mapping system combined with semantic segmentation deep learning to automate road inventory tasks. Their approach allows for precise extraction of road parameters, including width, lane configurations, superelevation, and others, using a 3D point cloud-based methodology. Compared to traditional methods, this solution reduces manual workload and enhances safety by minimizing the need for field inspections.

Expanding on automated road mapping, Cira et al. propose a deep learning-driven end-to-end road extraction system. Their method, applied to aerial orthophotography, utilizes semantic segmentation and conditional generative learning to refine road mapping accuracy at a state-level scale. By incorporating post-processing steps that enhance classification outputs, their work demonstrates how AI can significantly improve the accuracy and consistency of road network mapping.

Bai et al. contribute another significant advancement by integrating remote sensing images with vehicle trajectory data to generate road networks. By fusing movement-based data with image-based segmentation, their approach corrects inconsistencies in road mapping and enhances road connectivity accuracy. This is particularly relevant for urban and suburban road networks, where traditional satellite-based mapping may struggle with occlusions from buildings or trees.

1.2. Crack Detection and Pavement Condition Assessment: AI and Image Processing for Safer Roads

Pavement defects such as cracks, potholes, and surface roughness are critical indicators of road deterioration, yet detecting and quantifying these issues remains a major operational challenge. Manual inspections are resource-intensive and subjective, making automated detection methods a crucial advancement. This Special Issue presents three papers related to these topic.

Specifically, Deng et al. introduce a hybrid deep learning approach for crack detection and segmentation. Their methodology combines YOLOv5 for real-time crack detection with a modified Res-UNet model for precise segmentation. The results demonstrate higher accuracy in detecting cracks under complex environmental conditions, outperforming benchmark algorithms.

De León et al. take a different approach by developing a region-based minimal path selection algorithm, leveraging Gabor filters for enhanced crack segmentation. Unlike black-box deep learning models, their approach provides greater interpretability and robustness, making it an attractive option for real-world implementation in infrastructure monitoring and crack detection.

Fares and Zayed provide a comprehensive review of pavement roughness inspection technologies, analyzing five decades of industry and academic advancements. Their study discusses how 3D imaging, UAVs, and synthetic aperture radar are emerging as superior alternatives to traditional roughness measurement methods, highlighting trends that will shape the future of road condition monitoring.

1.3. Subsurface Imaging: GPR and AI for Underground Road Assessment

While surface-level defects are visible and easier to track, subsurface issues such as voids, material degradation, and water infiltration pose hidden risks to road pavement structural performance. This Special Issue features a study that integrates deep learning with GPR to enhance subsurface road condition assessment.

Li et al. propose an AI-enhanced clutter removal framework for GPR-based road assessment. Their approach incorporates contextual feature fusion and enhanced spatial attention modules, allowing for more precise subsurface imaging. The improved clutter removal method significantly enhances signal clarity and defect detection accuracy, making GPR a more viable tool for large-scale underground road analysis.

By combining deep learning with high-resolution GPR, this study illustrates how AI can improve subsurface imaging, helping road authorities detect structural weaknesses before they lead to costly failures.

1.4. Emerging Technologies in Road Monitoring and Traffic Infrastructure

Beyond mapping and defect detection, new sensor-based technologies are enabling enhanced road condition and safety and traffic infrastructure monitoring.

Lee and Cho apply AI-based image processing to assess pavement lane-marking visibility and retro-reflectivity. Their vehicle-mounted retro-reflectometer automates the collection of lane-marking data, providing a data-driven framework for prioritizing maintenance. This study is particularly relevant for urban environments where clear road markings are essential for ensuring satisfactory traffic safety.

Pan et al. focus on long-range perception systems using millimeter-wave radar, primarily for rail transit applications. Their approach enhances the real-time detection of road boundaries and obstacles, demonstrating its potential for autonomous vehicle navigation and traffic management.

Workman et al. propose a machine learning model for predicting unpaved road conditions using satellite imagery. Their approach is particularly useful for developing countries, where traditional road surveys are logistically challenging and expensive.

2. Specifics of Contributions

This section presents a structured analysis of the ten papers published in the Special Issue. To facilitate a comprehensive comparison, the contributions are categorized across four key aspects:

• Purposes and technologies;
• Study areas and data sources;
• Performance metrics and key indicators;
• Potential impacts on infrastructure management.

Table 1 summarizes the purposes, technologies exploited (i.e., the typology of surveys), and AI models or algorithms applied in each study.

Table 1. Purposes and technologies.

Paper	Purpose	Technology	AI Model/Algorithm
[Contribution 1]	Automated road inventory	Mobile mapping system	Semantic segmentation DL
[Contribution 2]	Crack detection and segmentation	Smartphone with Gimbal	YOLOv5 + Res-UNet algorithms
[Contribution 3]	Subsurface imaging	Ground Penetrating Radar	DL with CFFM and ESAM + RTM imaging
[Contribution 4]	State-level road mapping	Not Applicable (public data have been used)	DL + CGL
[Contribution 5]	Crack detection and segmentation	Not Applicable (public data have been used)	Dijkstra algorithm + Gabor Filters + RB-MPS algorithm
[Contribution 6]	Pavement roughness assessment	Not Applicable	Literature Review paper
[Contribution 7]	Long-range sensing of infrastructures	High-res long-range Millimeter-Wave Radar	Adapted DBSCAN method
[Contribution 8]	Unpaved road condition prediction	High-res optical satellite imagery	5 CNNs, VGG16 Network, and 15 supervised ML algorithms, and a blended model
[Contribution 9]	Road network generation	Remote sensing imagery + GPS (Trajectories)	SPBAM-LinkNet Algorithm
[Contribution 10]	Pavement lane marking assessment	Vehicle-mounted Retroreflectometer, high-res camera	Mask R-CNN and Otsu Threshold)

Acronyms: DL = deep learning, YOLO = You Only Look Once, Res-UNet = residual unity networking, CFFM = contextual feature fusion module, ESAM = enhanced spatial attention module, RTM = reverse time migration, CGL = conditional generative learning, RB-MPS = region-based minimal path selection, DBSCAN = density-based spatial clustering of applications with noise, CNN = convolutional neural networks, VGG16 = visual geometry group 16, SPBAM-LinkNet = split attention block and bottleneck attention module on the basis of LinkNet, R-CNN = regions with CNN.

As shown in Table 1, the studies employ a diverse range of methodologies and technologies to tackle different aspects of transport infrastructure detection, monitoring, and maintenance. AI-driven approaches dominate, particularly deep learning models, which are used for segmentation, classification, and data fusion. Crack detection methodologies also rely heavily on deep learning and advanced image processing. The methodologies summarized in Table 1 highlight the integration of AI, remote sensing, and sensor technologies as key drivers of innovation in infrastructure monitoring and maintenance.

Table 2 presents the geographical areas, and the type of data sources used in each study.

Table 2. Study areas and data sources.

Paper	Study Area	Data Source
[Contribution 1]	6 km of roads in Avila, Spain	3D point cloud data
[Contribution 2]	Public datasets have been used + testing images taken in a not-specified area	Road damage detection dataset, public road crack datasets, images from Hunan University, smartphone camera
[Contribution 3]	A road stretch in China	GPR survey data (synthetic + field data)
[Contribution 4]	Public datasets have been used	SROADEX Dataset (Aerial orthophotography)
[Contribution 5]	Public datasets have been used	DeepCrack dataset
[Contribution 6]	Global review (multi-continent)	Literature (Scopus and TRID Databases)
[Contribution 7]	Rail networks in China	Environmental point cloud data
[Contribution 8]	132 km of unpaved roads in Tanzania	CNES © 2017 AIRBUS DS (Optical satellite imagery)
[Contribution 9]	Rural roads, urban roads, expressways, and railways from six cities in China	CHN6-CUG and HB road datasets
[Contribution 10]	7293 km of freeways, arterials, and collector roads of Seoul, South Korea	Retroreflectometer data, high-res camera images

The scope of the studies varies significantly in terms of geographical focus and data sources, as illustrated in Table 2. Some studies focus on urban environments, while others assess rural and unpaved road networks. Three studies consider public datasets to enhance objectivity and transparency of their outcomes. As shown in Table 2, these studies collectively illustrate how diverse data sources are being used to improve road and railway management globally.

Table 3 highlights the evaluation metrics used to assess the effectiveness of each methodology.

Table 3. Performance metrics and key indicators.

Paper	Performance Metrics/Indicators
[Contribution 1]	IoU Score, Precision, Recall, F1-Score
[Contribution 2]	mAP, IoU, Pixel Accuracy, DICE, RMSE
[Contribution 3]	PNSR, SSIM
[Contribution 4]	Loss value, Accuracy, Precision, Recall, F1-Score, AUROC, IoU Score
[Contribution 5]	Precision, Recall, F1-Score
[Contribution 6]	Not applicable
[Contribution 7]	Object Detection Accuracy (X and Y position errors)
[Contribution 8]	Accuracy, Precision, Recall, F1-Score, FPR
[Contribution 9]	IoU Score, F1-Score, Precision–Recall Curve
[Contribution 10]	Markings are evaluated through retroreflectivity vs. ratio-of-defect relationship

Acronyms: IoU Score = Intersection over Union, mAP = mean Average Precision, DICE, Dice coefficient, RMSE = Root Mean Square Error, PSNR = mean Peak Signal-to-Noise Ration, SSIM = Structural Similarity Index Measure, AUROC = Area Under the Receiver operating characteristic, FPR = False Positive Rate.

Each study employs a specific set of performance metrics to validate its proposed methodology, as summarized in Table 3. Studies involving AI-based segmentation and object detection rely heavily on IoU Score, Precision, Recall, F1-Score, and Accuracy, which are very common numerical indicators in the assessment of data-driven classification algorithms.

Table 4 shows the potential impact of findings.

Table 4. Potential impacts on infrastructure management.

Paper	Potential Impact on Infrastructure Management
[Contribution 1]	It may improve efficiency and reduce manual inspections through AI-based road inventory mapping.
[Contribution 2]	It may enhance pavement-condition monitoring through novel deep learning methods.
[Contribution 3]	It may enable novel subsurface road assessments through AI-enhanced GPR imaging.
[Contribution 4]	It may automate large-scale infrastructure planning through novel AI-driven road mapping.
[Contribution 5]	It may offer an interpretable, non-black-box, low-cost, and robust crack detection method for road pavement management.
[Contribution 6]	It may guide future research through a comprehensive analysis of trends in pavement roughness assessment.
[Contribution 7]	It may improve rail transit serviceability and safety through innovative radar-based object detection.
[Contribution 8]	It may reduce the need for costly ground surveys, especially for unpaved roads in developing countries.
[Contribution 9]	It may enhance urban planning through a novel road network generation method, efficiently integrating GPS trajectory data and remote sensing images.
[Contribution 10]	It may allow prioritizing maintenance for road markings through novel AI-driven analysis.

As outlined in Table 4, the studies offer transformative contributions to road and railways management and maintenance. These studies collectively advance the field by enhancing automation, improving detection accuracy, and expanding the scope of monitoring solutions.

3. Conclusions

The research presented in this Special Issue, “Road Detection, Monitoring, and Maintenance Using Remotely Sensed Data”, highlights the transformative potential of remote sensing, AI-driven analysis, and advanced imaging techniques for infrastructure management. The ten studies featured in this issue showcase diverse methodologies that collectively improve the efficiency, accuracy, and scalability of monitoring and maintenance efforts.

By leveraging deep learning, machine learning, mobile mapping systems, AI-powered image processing, and novel non-black-box solutions, these studies address key challenges in road inventory, crack detection, pavement condition assessment, and road serviceability management.

It has been demonstrated that the adoption of NDT methods, remote sensing-based classification models, and data fusion techniques enables more effective decision-making for road and railway authorities, reducing manual labor, enhancing safety, and optimizing maintenance strategies.

4. The Second Special Issue: Continuing Scientific Advancements

Recognizing the success of this first edition, a second Special Issue titled “Road Detection, Monitoring, and Maintenance Using Remotely Sensed Data (2nd Edition)” has been launched. This follow-up Special Issue aims to expand on the research presented into the first edition, welcoming new contributions that further explore the intersection of remote sensing, AI, and road infrastructure management. Researchers and practitioners are encouraged to submit their latest findings to this second edition, available at: https://www.mdpi.com/journal/remotesensing/special_issues/243L1YZ43T, accessed on 19 February 2025).