Enhancing Concrete Crack Detection with YOLOV10-ViT Framework
Author Information
Author(s): Mayya Ali Mahmoud, Alkayem Nizar Faisal
Primary Institution: Nanjing University of Posts and Telecommunications, China
Hypothesis
Can a multi-stage deep learning framework improve the detection and classification of concrete cracks?
Conclusion
The proposed multi-stage YOLOV10-ViT model significantly enhances the accuracy of concrete crack classification compared to individual models.
Supporting Evidence
- The YOLOV10 model achieved a precision of 90.67%, recall of 90.03%, and F1-score of 90.34%.
- The multi-stage model outperformed the individual ViT model by 10.9%, 19.99%, and 19.2% for precision, recall, and F1-score, respectively.
- The ViT model reached a validation accuracy of 99.29% after training.
Takeaway
This study created a smart system that helps find and identify cracks in concrete, making buildings safer.
Methodology
The study used a multi-stage framework combining YOLOV10 for detection and ViT for classification of concrete cracks.
Limitations
The model may misclassify certain crack types due to resizing issues between detection and classification stages.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website