© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.

org (ISSN-2349-5162)

A Review of Pothole Detection Using Different

Technologies
1
Ashwini Jarali, 2Snehal Kavathekar, 3Sanskruti Lad, 4Prajwal Lalpotu,
5
Shreya Jadhav
1
Assistant Professor, 2Student, 3Student, 4Student, 5Student
1
Department of Computer Engineering,
1
International Institute of Information Technology, Pune, India

Abstract : This paper reviews recent advancements in pothole detection technologies, comparing various methods including
deep learning models like YOLO and SSD, and UAV-based systems with multispectral imaging. YOLO v8 Nano emerges as a
highly effective model, balancing speed and accuracy in real-time detection, while SSD demonstrates superior precision in certain
scenarios. Additionally, UAVs enhance detection by providing early insights into asphalt damage. Image processing techniques and
manually labelled datasets are also employed to improve model training and accuracy. The paper evaluates the strengths and
limitations of these methods, examining factors like computational efficiency, environmental adaptability, and real-time application.
It further explores future directions in this field, focusing on optimizing detection techniques and integrating advanced sensors to
enhance road safety and maintenance.

IndexTerms – YOLO(You Look Only Once), Pothole Detection

I. INTRODUCTION
The rapid deterioration of road infrastructure due to heavy traffic and environmental factors has highlighted the urgent need for
effective pothole detection and repair systems. Potholes, in particular, pose significant risks to vehicles and pedestrians, leading to
accidents, vehicle damage, and increased maintenance costs. Manual inspection, which has traditionally been the standard method
for identifying these hazards, is slow, labor-intensive, and subject to human error. Moreover, manual processes lack the scalability
and efficiency required for large-scale monitoring and real-time reporting. Automated pothole detection systems have emerged as a
promising solution to these challenges, leveraging advanced technologies like machine vision, sensor integration, and machine
learning. Deep learning models, such as Convolutional Neural Networks (CNNs), YOLO (You Only Look Once), and SSD (Single
Shot Detector), have demonstrated substantial improvements in detection speed and accuracy. These models are capable of processing
large volumes of data and identifying potholes in real time, thus enabling proactive road maintenance. Furthermore, UAV-based
systems equipped with multispectral imaging provide aerial surveillance, allowing for early and accurate detection of road surface
damage. This combination of technologies is transforming the road maintenance industry by offering scalable and efficient solutions.

II. AIM OF STUDY

The primary aim of this study is to conduct a detailed review of the current advancements in pothole detection technologies,
focusing on how deep learning models and UAV-based systems contribute to this field. By analyzing various methodologies,
including image processing techniques and machine learning algorithms, this review seeks to uncover the most effective strategies
for detecting potholes with high precision and efficiency. The research also aims to highlight how these technologies can be adapted
for practical use, with a focus on balancing computational efficiency, environmental robustness, and real-time applicability. The
paper aims to contribute to the ongoing development of automated road maintenance systems by providing insights into the strengths
and weaknesses of current technologies. Moreover, it addresses the need for further innovation in this field to overcome existing
limitations, such as the environmental sensitivity of sensors and the high computational demands of certain models. By identifying
gaps in the research, this study lays the groundwork for future improvements in pothole detection systems.
III. OBJECTIVE
The key objectives of this review are:
1. To explore and evaluate the variety of machine learning and sensor-based approaches utilized in pothole detection,
particularly focusing on advanced models like YOLO, SSD, and UAV-based multispectral imaging systems.

JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h11
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

2. To compare these methodologies in terms of their detection accuracy, speed, and ability to function in real-world conditions.
Particular attention will be given to their performance in varying environmental scenarios such as different lighting
conditions, weather effects, and road surface types.
3. To examine the common challenges faced by current detection systems, including issues related to the quality of the datasets
used for training models, the adaptability of algorithms across different environments, and the hardware constraints that
limit real-time deployment.
4. To propose potential future research directions that could improve the efficacy of these systems. This includes integrating
more advanced sensor technology, optimizing deep learning models for better generalization, and reducing the
computational and energy demands of these systems to enhance their scalability and usability.

IV. LITERATURE SURVEY

Pothole detection technology has evolved significantly over the past few decades, reflecting the continuous innovations aimed at
enhancing road safety and maintenance efficiency. Researchers have explored a wide range of methodologies, transitioning from
traditional sensor-based approaches to advanced deep learning algorithms, such as YOLO (You Only Look Once), and leveraging
modern imaging techniques like multispectral imaging with UAVs (Unmanned Aerial Vehicles). This literature review explores the
trajectory of pothole detection systems, beginning with the more conventional methods and progressing to the state-of-the-art,
vision-based techniques.
1. Traditional Approaches to Pothole Detection
Historically, pothole detection relied heavily on sensor-based techniques, particularly vibration and laser-based methods. These
systems typically focused on 3D reconstruction of road surfaces, using laser scanners and vibration sensors to detect road anomalies.
While Koch and Brilakis pioneered image-based methods for identifying road faults, subsequent research [1], such as Kang and
Choi's integration of 2D LiDAR and cameras [2], advanced the field by providing more detailed surface profiles. However, these
methods were limited by high implementation costs and challenges in scalability, especially for large-scale infrastructure projects
Vibration-Based Detection: Early approaches in pothole detection used accelerometers and vibration sensors installed on vehicles
to measure the force exerted by road irregularities [3]. The main drawback of vibration-based methods was their dependency on
vehicle speed and type, as well as the inability to provide precise locations or classifications of potholes.3D Reconstruction and
Laser Scanning: Another traditional method involved the use of laser scanning to create 3D reconstructions of road surfaces. These
systems provided detailed models of road topography, allowing for accurate identification of cracks and potholes [4]. However, the
high cost of equipment and the complexity of processing large amounts of data hindered widespread adoption. Kocha et al. (2022)
explored the use of both 2D visual recognition and 3D reconstruction for pothole measurement [5], demonstrating a method that
enhances accuracy while addressing some limitations. Though these approaches contributed to the early detection of road surface
abnormalities, their limitations spurred the exploration of alternative technologies that could overcome the issues of cost,
complexity, and scalability.
2. The Shift to Vision-Based Detection with CNNs
The advent of deep learning, particularly Convolutional Neural Networks (CNNs), marked a significant shift in pothole detection.
CNNs, known for their ability to learn and detect complex patterns in images, brought a new level of accuracy and efficiency to
road surface monitoring. Vision-based approaches, using both traditional RGB cameras and advanced imaging systems, proved to
be more affordable and scalable, although they initially faced challenges in terms of consistency under varying conditions, such as
lighting and weather [6]. Deep Learning and CNNs: A pivotal study, Deep Learning based Detection of Potholes in Indian Roads
using YOLO(2020), showcased the power of CNNs in detecting potholes in challenging environments [7]. The research highlighted
the use of the YOLO algorithm to process a dataset of potholes from Indian roads, which presented unique detection challenges due
to the diverse and often severe road conditions. The results demonstrated significant improvements over traditional methods,
particularly in terms of accuracy and processing speed, making vision-based techniques a more viable solution for large-scale
implementation [7].
3. YOLOv8: A Breakthrough in Real-Time Pothole Detection
As the need for faster and more accurate detection grew, the YOLO family of algorithms became the primary tool for vision-
based pothole detection. The YOLO architecture, designed for real-time object detection, allowed researchers to build models
capable of detecting potholes from video footage and images with minimal latency. In particular, the Pothole Detection System
Using YOLOv8 Algorithm (2024) presented a novel system for detecting potholes in real-time through live video feeds [8]. The
system was designed with a user-friendly Tkinter-based GUI, allowing users to upload images or videos for detection and providing
real-time alerts for road hazards. YOLOv8’s advanced object detection capabilities allowed the model to achieve high precision and
recall scores (0.95 and 0.97, respectively), making it a significant improvement over previous iteration like YOLOv3 and YOLOv4
[8]. The system’s accuracy and efficiency represent a leap forward in pothole detection, particularly for real-time applications in
urban environments.
4. UAV-Based Detection and Multispectral Imaging
While deep learning techniques like YOLO advanced pothole detection, researchers have also explored the integration of other
technologies, such as UAVs and multispectral imaging, to further improve accuracy and coverage. UAV Integration: In recent years,
UAVs have become a promising tool for road condition monitoring. Yifan Pan and Xianfeng Zhang’s research explored the use of
UAVs equipped with multispectral imaging devices to capture detailed surface data over large areas [9]. Unlike traditional RGB
imaging, multispectral imaging records spectral data at multiple wavelengths, enabling more precise detection of road defects. This
approach allowed researchers to capture more detailed information about pavement anomalies like potholes and cracks. The study
used machine learning models like Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF) to
classify road conditions based on the extracted spectral and textural data. Among these models, Random Forest (RF) outperformed
the others, achieving an overall accuracy of 98.3 [9].

JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h12
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

5. Comparative Analyses of Detection Algorithms

Numerous comparative studies have examined the trade-offs between different object detection algorithms. For example, Dhingra
et al. compared YOLO with other CNN-based models, such as Single Shot Detector (SSD) [10]. While SSD offered higher accuracy
in detecting smaller objects, YOLO was found to be significantly faster, making it ideal for real-time applications where immediate
feedback is critical, such as road safety systems. The ability to balance speed and accuracy remains a key consideration in the
development of pothole detection systems. While YOLO models are known for their rapid processing, other models, such as SSD,
may offer better performance in specific scenarios where detection of small objects is paramount. However, the scalability and real-
time capabilities of YOLO models make them more practical for widespread adoption in real-world applications. In summary,
pothole detection systems have progressed from early sensor-based approaches to modern, AI-driven solutions. The use of deep
learning algorithms like YOLOv8 [8], combined with advanced imaging techniques and UAVs [9], has greatly improved the speed,
accuracy, and scalability of these systems. As research continues, the integration of additional data sources and the development of
more sophisticated algorithms will likely lead to even more effective pothole detection and road maintenance systems, contributing
to enhanced road safety and infrastructure longevity.

V. CHALLENGES AND LIMITATIONS

Despite the advancements in pothole detection technology, several challenges and limitations persist. These challenges can affect
the accuracy, efficiency, and implementation of detection systems, hindering their overall effectiveness in improving road safety
and infrastructure maintenance.
• Limitations of Traditional Approaches:
Traditional pothole detection methods, such as sensor-based techniques and manual inspections, have contributed to identifying
road defects. However, these approaches often struggle to provide timely and accurate assessments, especially in dynamic
environments. There is a pressing need for advanced methodologies that leverage deep learning to enhance the precision and
efficiency of pothole detection, particularly in real-time scenarios.
• Environmental Factors:
Variable Lighting Conditions: One of the primary challenges faced by vision-based detection systems is their reliance on
consistent lighting conditions. Pothole detection algorithms, particularly those based on CNNs like YOLO, can struggle with
shadows, reflections, and glare, which may lead to false positives or missed detections. Weather Conditions: Adverse weather
conditions such as rain, snow, or fog can significantly impede the performance of imaging systems. For example, heavy rain
may obscure potholes and reduce visibility, while snow can cover road defects, making detection difficult.
Road Surface Characteristics: Variability in road surface textures and materials can also affect detection accuracy. Different
types of pavements may reflect light differently or have unique surface features that challenge the algorithms’ ability to identify
potholes consistently.
• Data Quality and Availability:
Dataset Limitations: The performance of machine learning models is heavily reliant on the quality and diversity of the training
datasets. Many existing datasets may not adequately represent the diverse range of potholes found in various geographical
regions, especially in developing countries. This limitation can lead to reduced accuracy when models are applied to unfamiliar
environments. Annotation Challenges: Accurate annotation of training data is crucial for supervised learning approaches. The
process of labelling images with potholes can be time consuming and prone to human error, affecting the model’s learning and
performance.
• Computational Resource Requirements:
Deep learning algorithms typically demand substantial computational resources for training and inference, especially in real-
time applications. The need for high performance hardware can limit the accessibility and scalability of these solutions,
particularly in resource constrained environments. Future research should explore optimizing model architectures to balance
performance and resource efficiency.
• Generalization and Robustness:
Overfitting: Deep learning models can become overfitted to the training data, leading to poor performance when exposed to new
or unseen conditions. Ensuring that models generalize well across different environments is a significant challenge.
Performance Variability: Variations in the performance of detection systems across different conditions (e.g., urban vs. rural
settings) can lead to inconsistent results, limiting their reliability as a comprehensive solution for pothole monitoring.
These challenges collectively underscore the importance of continued research to refine deep learning methodologies, improve
model robustness, and ultimately enhance the reliability of pothole detection systems. By addressing these limitations, the
integration of deep learning into pothole detection can significantly contribute to safer roads and more efficient infrastructure
maintenance.

VI. DATASET DESCRIPTION

This study utilizes a pothole detection dataset generated from Roboflow Universe. The dataset consists of 665 annotated images,
which have been carefully labelled to identify potholes in various road conditions. Out of the total dataset, 465 images are used for
training the model, while 67 images are set aside for testing, ensuring a robust evaluation of the model’s performance.
The dataset is well-suited for real-time object detection tasks and includes various annotations to capture the diversity of potholes
in different environments. The images are pre-processed and formatted to be compatible with YOLOv5 and YOLOv8 models. These
models are trained and deployed using the Roboflow API, which allows seamless integration and application in the system for
automated pothole detection. Key details of the dataset includes:

JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h13
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Training Images: 465

•
Testing Images: 67
•
• Annotation Format: COCO JSON
• Label Types: Bounding boxes indicating pothole locations

This dataset is optimized for deep learning applications, and the use of the Roboflow API facilitates efficient model training,
deployment, and continuous improvement as more data becomes available. By leveraging this dataset, the system aims to detect
potholes in real-world environments with high accuracy and speed, contributing to enhanced road safety and maintenance.

VII. RESEARCH METHODOLOGY

This section outlines the approach used to preprocess pothole detection data, build deep learning models, and train them to identify
potholes from road surface images. The methodology includes steps from data acquisition to model evaluation, ensuring robustness
and accuracy.

1. Data Preprocessing
The dataset used for training the pothole detection model was sourced from Roboflow, containing 665 annotated images of road
surfaces with and without potholes. Key preprocessing steps include:
• Loading the Dataset: The dataset was loaded and organized using the Roboflow API to streamline access to annotated images.
• Resizing and Augmentation: Images were resized to 640x640 pixels for YOLOv8 compatibility, and augmentation techniques
like flipping, rotation, and brightness adjustment were applied to increase model generalization.

Figure1. Training and Testing Data

• Bounding Box Annotations: Each image was annotated with bounding boxes around potholes. The dataset adhered to the
COCO format for object detection, facilitating seamless integration with the YOLOv8 model.
• Train-Test Split: The dataset was split into 80% training and 20% testing data to ensure robust model evaluation.

2. Model Building and Training

Both YOLOv8 and YOLOv5 are employed for pothole detection. These models are chosen for their ability to balance speed and
accuracy. The following steps describe the model development:
• Model Architecture: YOLOv5 and YOLOv8 utilize a single-stage detection mechanism where the network predicts bounding
boxes and class probabilities in one pass.
The loss function is a combination of three components:
– Classification Loss: Measures the error in predicting the class of detected objects.
– Localization Loss (2 norm): The difference between the predicted bounding box and the ground truth bounding box .
– Confidence Loss: Evaluates how well the model predicts the presence of an object within the bounding box.
• Transfer Learning: Pre-trained weights from the COCO dataset are used to initialize both YOLOv5 and YOLOv8 models. This
accelerates convergence and improves accuracy.
• Training: The models are trained using a Stochastic Gradient Descent (SGD) optimizer with the following loss function:

𝐿𝑡𝑜𝑡𝑎𝑙 = 𝐿𝑐𝑙𝑎𝑠𝑠 + 𝐿𝑏𝑜𝑥 + 𝐿𝑐𝑜𝑛𝑓 (1)

where 𝐿𝑐𝑙𝑎𝑠𝑠 is the classification loss, 𝐿𝑏𝑜𝑥 is the bounding box regression loss, and 𝐿𝑐𝑜𝑛𝑓 is the confidence loss. Hyperparameter
Tuning: Learning rate 𝛼, batch size, and momentum (m) are tuned to optimize performance. The learning rate decay formula used
during training is:

1
𝛼 𝑡 = 𝛼0 ( ) (2)
1+𝑡
JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h14
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

where 𝛼0 is the initial learning rate, is the decay factor and 𝑡 is the current epoch.

3. Performance Evaluation :
The performance of the YOLO models is evaluated using the following metrics:
• Precision and Recall: Precision measures the proportion of correct pothole detections out of all detections made:

𝑇𝑃
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (3)
𝑇𝑃 + 𝐹𝑃

where TP is true positives and FP is false positives. Recall measures the proportion of actual potholes detected by the model:

𝑇𝑃
𝑅𝑒𝑐𝑎𝑙𝑙 = (4)
𝑇𝑃 + 𝐹𝑁

where FN is false negatives.

• Mean Average Precision (mAP): 𝑚𝐴𝑃 is used to evaluate the model’s detection accuracy across multiple thresholds. It is the
average precision across various Intersection over Union (IoU) thresholds:
𝑛
1
𝑚𝐴𝑃 = ∑ 𝐴𝑃𝑖 (5)
𝑛
𝑖=1

where 𝐴𝑃𝑖 is the average precision for class 𝑖 and 𝑛 is the number of classes.

VIII. RESULTS AND DISCUSSION

This section evaluates the performance of YOLOv5 and YOLOv8 models for pothole detection, focusing on precision, recall,
and mean Average Precision (mAP) as the primary metrics. The loss values for object detection and classification during training
and validation are also examined to assess the models’ ability to learn effectively. The results for the YOLOv5 and YOLOv8 models
are presented in Table 1, which compares the precision, recall, and mAP scores at different Intersection over Union (IoU) thresholds.

Table 1
Performance Comparison of Yolov5 and Yolov8 Models
Model Precision Recall mAP@0.5 mAP@0.5:0.95
Yolov5 0.392 0.239 0.251 0.0878
Yolov8 0.726 0.606 0.697 0.425

The results in Table 1 shows that YOLOv8 outperforms YOLOv5 across all performance metrics. YOLOv8 achieves a higher
precision (0.72) compared to YOLOv5 (0.39), indicating fewer false positives. The recall for YOLOv8 is also higher (0.60) than
YOLOv5 (0.23), meaning that YOLOv8 is better at detecting potholes correctly without missing as many true positives.
The mAP@0.5 score, which reflects the model’s ability to correctly detect potholes with an IoU threshold of 0.5, is significantly
higher for YOLOv8 (0.69) compared to YOLOv5 (0.25). Similarly, the mAP@0.5:0.95, which averages IoU thresholds from 0.5 to
0.95, also shows YOLOv8 leading with a score of 0.87 versus YOLOv5’s score of 0.42. These results indicate that YOLOv8
provides better overall detection accuracy across various IoU thresholds.
Loss Curves: The training and validation loss curves for box loss, objectness loss, and classification loss are crucial for
understanding model learning behavior. Both models show a steady decline in these losses over the course of training, with YOLOv8
demonstrating a faster convergence, indicating more efficient learning. This is supported by the reduced validation losses in
YOLOv8, suggesting that it generalizes better to unseen data. In the train/seg_loss and train/dfl loss_graphs, YOLOv8 demonstrates
better segmentation performance with a lower loss throughout the epochs, further solidifying its superiority over YOLOv5 in
handling detailed object segmentation tasks such as pothole detection. The overall results demonstrate that YOLOv8 outperforms
YOLOv5 across key metrics, making it the better choice for pothole detection. The improvements in precision and recall suggest
that YOLOv8 reduces false positives and false negatives more effectively. Its higher mAP scores at varying IoU thresholds indicate
that YOLOv8 can detect potholes more reliably, even when stricter localization is required.

JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h15
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Figure2. Training and Validation Loss Curves for YOLOv8

Figure3. Training and Validation Loss Curves for YOLOv5

Despite these advancements, there remains room for improvement. The mAP@0.5:0.95 scores for both models show that the
performance still drops as the IoU threshold becomes more stringent, indicating that the models could benefit from further
optimization, such as hyperparameter tuning or additional data preprocessing steps to enhance object localization. Future research
could explore the integration of temporal data for tracking potholes over time, the use of higher resolution imagery for improved
detection accuracy, and additional refinements in model architecture or training strategies.

IX. CONCLUSION
In this paper, we have explored the evolution of pothole detection systems, highlighting the shift from traditional sensor-based
methods to modern, deep learning-based approaches. The advancements in technologies like YOLOv8 and UAV integrated
multispectral imaging have greatly enhanced the accuracy, speed, and scalability of detection systems. YOLOv8, in particular, has
demonstrated impressive performance in real time detection, making it a practical solution for large-scale applications. Despite
these advancements, challenges such as environmental factors, dataset limitations, and computational resource demands persist.
Future research should focus on optimizing model architectures, improving dataset diversity, and addressing generalization issues
to further enhance the effectiveness of pothole detection systems, contributing to safer roads and more efficient infrastructure
maintenance.

REFERENCES
[1] Koch, Christian & Brilakis, Ioannis. (2011). Pothole detection in asphalt pavement images. Advanced Engineering Informatics.
25. 507-515. 10.1016/j.aei.2011.01.002.
[2] B. -h. Kang and S. -i. Choi, "Pothole detection system using 2D LiDAR and camera," 2017 Ninth International Conference on
Ubiquitous and Future Networks (ICUFN), Milan, Italy, 2017, pp. 744-746, doi: 10.1109/ICUFN.2017.7993890.
[3] C. Chellaswamy, H. Famitha, T. Anusuya and S. B. Amirthavarshini, "IoT Based Humps and Pothole Detection on Roads and
Information Sharing," 2018 International Conference on Computation of Power, Energy, Information and Communication
(ICCPEIC), Chennai, India, 2018, pp. 084-090, doi: 10.1109/ICCPEIC.2018.8525196
[4] A. Ahmed, M. Ashfaque, M. U. Ulhaq, S. Mathavan, K. Kamal and M. Rahman, "Pothole 3D Reconstruction With a Novel
Imaging System and Structure From Motion Techniques," in IEEE Transactions on Intelligent Transportation Systems, vol. 23,
no. 5, pp. 4685-4694, May 2022, doi: 10.1109/TITS.2021.3054026.
JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h16
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

[5] Jog, Gauri & Koch, Christian & Golparvar-Fard, Mani & Brilakis, Ioannis. (2012). Pothole Properties Measurement through
Visual 2D Recognition and 3D Reconstruction. 10.1061/9780784412343.0070.
[6] Safyari Y, Mahdianpari M, Shiri H. A Review of Vision-Based Pothole Detection Methods Using Computer Vision and Machine
Learning. Sensors (Basel). 2024 Aug 30;24(17):5652. doi: 10.3390/s24175652.
[7] D. J, S. D. V, A. S A, K. R and L. Parameswaran, "Deep Learning based Detection of potholes in Indian roads using
YOLO," 2020 International Conference on Inventive Computation Technologies (ICICT), Coimbatore, India, 2020, pp. 381-
385, doi: 10.1109/ICICT48043.2020.9112424.
[8] S. Sasane, A. Tickoo, K. Xavier, and M. Parmar, "Pothole detection system using YOLO v8 algorithm," International Journal
of Creative Research Thoughts (IJCRT), vol. 12, June 2024.
[9] Y. Pan, X. Zhang, G. Cervone and L. Yang, "Detection of Asphalt Pavement Potholes and Cracks Based on the Unmanned
Aerial Vehicle Multispectral Imagery," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,
vol. 11, no. 10, pp. 3701-3712, Oct. 2018, doi: 10.1109/JSTARS.2018.2865528
[10] Dhingra, Mayank & Dhingra, Rahul & Sharma, Meghna. (2024). Pothole Detection Using Machine Learning Models.
International Journal of Scientific Research in Science, Engineering and Technology. 11. 94-105. 10.32628/IJSRSET241126.

JETIR2412702 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org h17