Related Experiment Video
Updated: Jun 1, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
462
Comparative analysis of deep learning models for crack detection in buildings.
S Siva Rama Krishnan1, M K Nalla Karuppan2, Adil O Khadidos3
1School of Computer Science and Information Systems, Vellore Institute of Technology, Katpadi, Vellore, 632014, Tamilnadu, India.
Scientific Reports
|January 17, 2025
Summary
This study introduces a novel deep learning approach for automated crack detection in brickwork, achieving 99.98% accuracy. The research addresses data scarcity and enhances building safety through precise crack identification.
Area of Science:
- Civil Engineering
- Computer Science
- Artificial Intelligence
Background:
- Building structures are susceptible to cracks from environmental factors and material degradation.
- Accurate assessment and maintenance are crucial for building safety and longevity.
- Automated crack detection using Artificial Intelligence (AI) offers efficient solutions for civil engineering challenges.
Purpose of the Study:
- To develop and curate a novel deep learning image processing method for detecting cracks in brickwork.
- To address the research gap and data scarcity in automated crack identification.
- To train and validate deep learning models for classifying brickwork images as cracked or normal.
Main Methods:
- A dataset of 24,000 brickwork images was curated and classified into crack and non-crack categories.
- Several deep learning models, including Inception V3, VGG-16, RESNET-50, VGG-19, Inception ResNetV2, and CNN-RES MLP, were trained and validated.
- Model performance was evaluated using parameters like Batch size, Pooling, Activation functions, Learning-rate, Kernel-Size, Normalization, and Optimizers.
Main Results:
- Inception V3 achieved the highest accuracy at 99.98%.
- Inception V3 demonstrated the best Precision at 99.99%.
- RESNET-50 achieved the highest Recall at 99.98%, and Inception V2 showed the best Region of Convergence (0.9999).
Conclusions:
- Deep learning models, particularly Inception V3, are highly effective for automated crack detection in brickwork.
- The developed approach addresses data scarcity and offers a reliable solution for identifying structural surface cracks.
- This research contributes to improved building safety and maintenance through precise and automated crack assessment.
