Optimized Pavement Crack Segmentation with Low Computational Cost Using Fusion-Enhanced Attention U-Net
Release time:2026-07-20
Hits:
- Affiliation of Author(s):
- 城市发展与现代交通学院
- Journal:
- Journal of Computing in Civil Engineering
- Abstract:
- Accurate segmentation of pavement cracks is crucial for maintaining road safety and the longevity of road infrastructures. Existing convolutional neural network (CNN) models often struggle with the noise and irregular crack patterns inherent in real-world pavement images. Although attention mechanisms have been introduced to address these challenges, they frequently result in high computational costs. In this paper, we propose an optimized pavement crack segmentation method using a fusion-enhanced Attention U-Net. This model integrates global and local semantic information through a novel fusion-enhanced attention mechanism, effectively highlighting critical regions in pavement images while maintaining computational efficiency. We extensively evaluated the proposed model on three publicly available data sets: DeepCrack, Crack500, and crack forest dataset (CFD). The model’s performance was assessed using a range of accuracy metrics, including Intersection over Union (IoU), F1-score, precision, recall, and accuracy. The model demonstrated superior performance compared with six state-of-the-art models: SegNet, fully convolutional network (FCN), U-Net, Attention U-Net, DeepLabv3+, and residual block with convolutional unit (ResBCU-Net). Furthermore, analyses of memory and time complexities indicated that our model offers reduced computational costs, demonstrating its practicality and effectiveness for pavement crack segmentation tasks and confirming its suitability for real-time infrastructure monitoring applications.
- First Author:
- MOHAMMEDMOHAMMEDAMEENABDULHAMEED
- Indexed by:
- Journal paper
- Volume:
- 40 / 4 / 04026041
- ISSN No.:
- 0887-3801
- Translation or Not:
- no
- Date of Publication:
- 2026-04-29