@article { author = {Padsumbiya, Mihir and Brahmbhatt, Vedant and Thakkar, Sonal}, title = {Automatic Crack Detection Using Convolutional Neural Network}, journal = {Journal of Soft Computing in Civil Engineering}, volume = {6}, number = {3}, pages = {1-17}, year = {2022}, publisher = {Pouyan Press}, issn = {2588-2872}, eissn = {2588-2872}, doi = {10.22115/scce.2022.325596.1397}, abstract = {Manual inspection of cracks on concrete surfaces requires wholesome knowledge and depends entirely on the expertise and capabilities of the inspector. This study proposes the use of a simple Convolutional Neural Network (CNN) for automatic crack detection. A comparative approach for Automated Crack Detection is presented between Feed-Forward Fully Connected Neural Networks and CNN, focusing on the primary hyperparameters affecting the accuracy of both systems. An inclination towards CNN is concluded due to its simplicity and computational efficiency. For the purpose of this study, the input data is extracted from an open-source platform. In the second step, the images are pre-processed for obtaining low-pixel density images with the aim to get better accuracy at lower computer power. The CNN proposed uses Max Pooling and appropriate optimization techniques. The model is trained to detect and segregate cracked and non-cracked concrete surfaces through input images. The proposed model predicts and labels images with cracks on concrete surfaces and images with no cracks using pixel-level information. The final accuracy achieved is 97.8% by the proposed CNN model. The proposed model is a novel approach to detecting cracks on low pixel density images of concrete surfaces for its economic and processing efficiency and thus eliminates the need for high-cost digital image capturing devices. This study signifies and confirms the impact of Artificial Intelligence in the Civil Engineering field where using simple techniques like a simple four-layered Neural Network is capable of carrying automatic inspection of cracks which can be further developed for other applications.}, keywords = {Deep Learning,computer vision,Maintenance and rehabilitation,Artificial Neural Network}, url = {https://www.jsoftcivil.com/article_153278.html}, eprint = {https://www.jsoftcivil.com/article_153278_2bf7e467d0f431a073b206e89cab394f.pdf} }