.png){kind=icon}
.png){kind=icon}
Return
Automated pixel-level detection of pavement cracks based on deep learning of multi-source image fusion
.png){kind=icon}
PRE
AI
J
X
K
H
J
R
DOI：10.1080/10298436.2026.2688421.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Detecting road cracks is essential for safety and long-term road maintenance. This paper improves the accuracy and robustness of pixel-level pavement crack detection by addressing limitations in existing neural networks using single-source images. It introduces multi-dimensional preprocessing techniques, including the iterative gradient suppression and refinement (IGSR) algorithm and noise removal methods for both grey and depth images, enhancing the data quality. This study combines 2D and 3D pavement data and uses multi-source image fusion to strengthen feature representation and reduce information loss during neural network extraction. The application of enhanced AI segmentation models further boosts detection performance. The experimental results show that fusion strategies improve the F1-score by 12.27% compared to grey image-based segmentation and by 5.65% compared to depth image-based segmentation. Overall, the method enhances crack detection precision and robustness, leading to more timely and accurate maintenance, prolonging infrastructure lifespan, and improving public safety in transportation.
Keywords:
Pixel-level crack detection
heterogeneous data
deep learning
data pre-processing
image fusion
artificial intelligence
Journal
IF:
3.3
Papers:
2.8K
Citations:
8.0K
Organization
Cited Papers