Concrete Crack and Spall Dataset for Segmentation and Classification

This dataset, which can be used for vision-based deep learning methods, can been implemented to detect and analyze damages in concrete structures. In order to increase the generalizability of the network results, a number of images from three datasets in references [1-3] were combined as follows which can be downloaded from https://www.kaggle.com/datasets/stmlen/cconcrack. (If you use this dataset, please cite this research paper: https://arxiv.org/abs/2501.11836) • One hundred Turkish images were randomly selected from the Özgenel segmentation dataset [1] with a resolution of 30244032. • One hundred Turkish images were randomly selected from the reference dataset [3] with different resolutions such as 296306 and 334306. • Two hundred crack and drop images were randomly selected from the reference dataset [2] with different resolutions such as 768768 and 960*1280. In general, 400 concrete cracks and spall images (2 categories) with different resolutions were randomly selected from the above references, which were then increased to 10,995 images containing concrete cracks and spalls. Also, all images were changed to 640 x 640 resolution with pre-processing. After selecting the desired datasets, we masked them with the Roboflow labeling tool for the semantic segmentation task. 1- Özgenel, Çağlar Fırat. "Concrete crack segmentation dataset." Mendeley Data 1 (2019). 2- Zhang, Chaobo, Chih-chen Chang, and Maziar Jamshidi. "Simultaneous pixel-level concrete defect detection and grouping using a fully convolutional model." Structural Health Monitoring 20.4 (2021): 2199-2215. 3- Shawn, Yang. "FCN for crack recogntion", 13 March 2018, github.com/OnionDoctor/FCN_for_crack_recognition