Deep learning-based crack detection and prediction for structural health monitoring
Abstract This study describes a system for real-time crack detection and estimation in constructions, using a proposed model based on CNN, VGG16, U-Net, and Swin Transformer architectures. This model classifies and segments structural cracks irrespective of their severity with 98.88% precision. Prep...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-06-01
|
| Series: | Discover Applied Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s42452-025-07272-y |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract This study describes a system for real-time crack detection and estimation in constructions, using a proposed model based on CNN, VGG16, U-Net, and Swin Transformer architectures. This model classifies and segments structural cracks irrespective of their severity with 98.88% precision. Preprocessing techniques such as oversampling and class under sampling mitigate imbalance, while augmentation aids generalization, was followed. Detection performance is improved by fine-tuning VGG16 using transfer learning. Robust segmentation and identification are provided by U-Net and Swin Transformer. A focus on a visual model indicates cracks and inspects the possibility of damage to the structure in the future by employing bounding boxes. Recurrent structure problems, validated by RGB histograms and confusion matrices, are justified and analyzed. Improving the precision detection of the system contributes to preventive maintenance, which significantly reduces the cost of repairs and increases safety. Using deep learning models within the framework enables effective and automated structural health monitoring. This approach allows for the monitoring of cracks, damage progression, and other relevant details and serves as a critical step toward infrastructure management, protection, and risk evaluation for disaster prevention and mitigation. |
|---|---|
| ISSN: | 3004-9261 |