Bridge Crack Detection Using Convolutional Neural Networks

Abstract

Every day, millions of people drive over bridges without thinking twice about what’s holding them up. But for civil engineers and infrastructure authorities, bridge health is a constant and serious concern. These structures age under relentless stress — traffic loads, temperature swings, corrosion, and moisture all chip away at the concrete over time. Cracking is usually the earliest visible sign that something is starting to go wrong, and catching it early is the difference between a quick repair and a major structural failure. Traditional inspection relies on engineers physically walking and examining bridges — a process that is slow, subjective, and sometimes dangerous. In this paper, we present a CNN-based automated crack detection system that learns to distinguish cracked from intact concrete surfaces directly from images. Trained on a large labeled dataset, our model achieved a classification accuracy of 93.22 percentage, showing that deep learning can meaningfully automate this part of structural inspection.

Introduction

Bridge infrastructure degrades over time, and cracks are one of the earliest indicators of structural damage. If ignored, small cracks can expand, allow water ingress, corrode reinforcement steel, and ultimately compromise bridge safety. Traditional inspection methods rely on manual visual checks, which are slow, costly, and prone to subjective errors. To address this, the study proposes an automated crack detection system using a Convolutional Neural Network (CNN) to identify cracks from bridge surface images.

The model is trained on the Kaggle Concrete Crack Image Dataset, which contains labeled images of cracked and non-cracked concrete under varying lighting and surface conditions. All images are preprocessed by resizing to 120×120 pixels and normalizing pixel values to ensure stable and efficient training.

The CNN learns hierarchical visual features through convolutional and pooling layers, starting from simple edges and textures to more complex crack patterns. The extracted features are then passed through fully connected layers, producing a binary output (crack or no crack) using a sigmoid activation function.

Experimental results show that the model is trained using an 80/20 style data split with 20 epochs and achieves a testing accuracy of about 93.22%. The system demonstrates stable learning behavior with improving accuracy and decreasing loss during training. Evaluation using a confusion matrix further helps analyze correct and incorrect predictions, confirming the model’s effectiveness for automated bridge crack detection.

Conclusion

This project set out to answer a practical question: can a CNN reliably detect cracks in bridge surface images? Based on our results, yes — at least at a level that makes the approach worth taking seriously. We built and trained a CNN on the Kaggle Concrete Crack dataset and achieved a classification accuracy of 93.22 percent on images the model had never seen. The confusion matrix shows the modelis particularly good at avoiding false alarms (only 8 false positives), though the 114 missed cracks suggest recall could be improved through a larger or more diverse dataset, or a deeper architecture. The system is not ready to replace experienced inspectors outright. Real-world bridge surfaces are messier than a benchmark dataset, and the consequences of a missed crack are serious. But as a first-pass screening tool — something that flags suspicious areas for an engineer to examine more closely — this kind of system has real and immediate potential. Going forward, we would like to explore transfer learning from pre-trained models like ResNet or VGG, which could push accuracy higher without requiring more training data. We are also interested in integrating this with drone-based capture systems for continuous, large-scale bridge monitoring.

References

[1] Y. J. Cha, W. Choi, and O. Buyukozturk, “Deep learning-based crack damage detection using convolutional neural networks,” Computer-Aided Civil and Infrastructure Engineering, vol. 32, no. 5, pp. 361–378, 2017. [2] S. Dorafshan, R. J. Thomas, and M. Maguire, “Comparison of deep convolutional neural networks and edge detectors for image-based crack detection in concrete,” Construction and Building Materials, vol. 186, pp. 1031–1045, 2018. [3] A. Krizhevsky, I. Sutskever, and G. Hinton, “ImageNet classification with deep convolutional neural networks,” Advances in Neural Information Processing Systems, 2012, pp. 1097–1105. [4] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in Proc. International Conference on Learning Representations (ICLR), 2015. [5] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778.

Copyright

Copyright © 2026 Kavish Gupta, Chander Dev Singh, Saksham Choudhary, Yash Jogdand, ANIL M. BHADGALE. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.