Techno Press
Techno Press

Advances in Concrete Construction
  Volume 18, Number 2, August 2024 (Special Issue) pages 147-155
DOI: https://doi.org/10.12989/acc.2024.18.2.147
 


Automatic crack detection using quantum-inspired firefly algorithm with deep learning techniques
K.A. Vinodhini, K.R. Aswin Sidhaarth and K.A. Varun Kumar

 
Abstract
    Detecting and quantifying cracks in bituminous (asphalt) road surfaces plays a crucial role in maintaining road infrastructure integrity and enabling cost-effective maintenance strategies. However, traditional manual inspections are laborious, time-intensive, and susceptible to inconsistencies due to factors like human fatigue, varying expertise levels, and subjective assessments. To address these challenges, this research proposes CrackNet, an innovative deep learning framework that harnesses state-of-the-art computer vision and object detection techniques for accurate and computationally efficient automated crack detection in bituminous road imagery. CrackNet introduces a novel hybrid neural network architecture that seamlessly integrates a cutting-edge Vision Transformer backbone with multi-scale convolutional feature fusion modules. The Vision Transformer component excels at capturing long-range structural dependencies and global contextual information, while the multi-scale fusion modules adeptly combine fine-grained crack details across various spatial resolutions. This unique design enables CrackNet to holistically model intricate crack topologies while preserving localized characteristics and intricate details. To further bolster robustness and generalization capabilities across diverse real-world scenarios, CrackNet incorporates selfsupervised pre-training techniques that leverage unlabeled data and unsupervised pretext tasks. These strategies allow CrackNet to learn rich visual representations tailored specifically for crack detection. Additionally, an extensive data augmentation pipeline is employed, encompassing geometric, photometric, and adversarial transformations, to enhance model invariance to varying imaging conditions and environmental factors. The accuracy achieved by the newly proposed approach surpasses that of current state-of-the-art methodologies, reaching an impressive 97.8%.
 
Key Words
    bituminous pavements; deep learning; multi-scale fusion; object detection; road crack detection; selfsupervised learning; vision transformers
 
Address
(1) K.A. Vinodhini:
Vel Tech High Tech Dr. Rangarajan Dr. Sakunthala Engineering College (An Autonomous Institution), India;
(2) K.R. Aswin Sidhaarth:
Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, India;
(3) K.A. Varun Kumar:
Department of Networking and Communications, School of Computing, SRM Institute of Science and Technology, Kattankulathur - 603203, India.
 
References
    -acc1802008-
 

Techno-Press: Publishers of international journals and conference proceedings.       Copyright © 2026 Techno Press
P.O. Box 33, Yuseong, Daejeon 305-600 Korea, Tel: +82-42-828-7996, Fax : +82-42-828-7997, Email: admin@techno-press.com