CrackGNet: A Hybrid Approach Leveraging U-Net and Graph Neural Networks for Road Crack Detection

aut.embargoYes
aut.embargo.date2026-08-14
dc.contributor.advisorYongchareon, Sira
dc.contributor.authorParas, Paras
dc.date.accessioned2024-08-13T23:16:34Z
dc.date.available2024-08-13T23:16:34Z
dc.date.issued2024
dc.description.abstractPavement cracks are a common issue affecting road infrastructure worldwide. Their timely detection and repair are essential for ensuring road safety and preventing further deterioration. Traditional manual crack inspection methods are time-consuming, labor-intensive, and susceptible to human mistakes. To overcome these challenges, this thesis introduces CrackGNet, a hybrid deep-learning architecture for the automated detection of road cracks. Our proposed method strategically integrates a U-Net encoder-decoder with a GCN. The U-Net component focuses on robust feature extraction from crack images, while the GCN component explicitly models the spatial relationships and connectivity patterns inherent in crack structures. We performed a comprehensive evaluation of CrackGNet on the CFD, DeepCrack, and CRACK500 datasets, demonstrating its superiority over established deep learning baselines across a range of performance metrics achieving 0.8618 precision, 0.8811 recall, and an F1 score of 0.8665, outperforming existing methods. These results highlight the effectiveness of our hybrid approach and the importance of considering both local image features and spatial context in crack detection tasks. Our research contributes to the advancement of automated road infrastructure assessment systems, with the potential to improve infrastructure maintenance efficiency and reduce costs. Future research directions include exploring alternative GNN architectures for further performance gains, expanding datasets for increased model robustness, and evaluating CrackGNet in real-world inspection scenarios.
dc.identifier.urihttp://hdl.handle.net/10292/17879
dc.language.isoen
dc.publisherAuckland University of Technology
dc.rights.accessrightsOpenAccess
dc.subjectRoad Crack Segmentation
dc.subjectDeep Learning
dc.subjectU-Net
dc.subjectGraph Neural Networks
dc.titleCrackGNet: A Hybrid Approach Leveraging U-Net and Graph Neural Networks for Road Crack Detection
dc.typeThesis
thesis.degree.grantorAuckland University of Technology
thesis.degree.nameMaster of Philosophy
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
paras_thesis.pdf
Size:
19.45 MB
Format:
Adobe Portable Document Format
Description:
Thesis is embargoed until 14 August 2026
License bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
license.txt
Size:
897 B
Format:
Item-specific license agreed upon to submission
Description:
Collections