New-TunnelCrack
收藏IEEE2026-04-17 收录
下载链接：
https://ieee-dataport.org/documents/new-tunnelcrack
下载链接
链接失效反馈官方服务：
资源简介：
The New-TunnelCrack dataset is constructed based on the publicly available LinkCrack dataset (https:\/\/github.com\/urban-informatics\/LinkCrack), which contains tunnel lining images with visible cracks.A total of 981 crack-containing images were selected as the initial dataset. To enhance model generalization and mitigate class imbalance, two data augmentation strategies\u2014HSV color-space transformation and random rotation\u2014were designed and applied to increase sample diversity.After augmentation, the dataset was expanded to 2,943 images. Crack regions were manually annotated using LabelImg, and corresponding bounding-box label files were generated.The final dataset was randomly divided into training (2,355 images), validation (294 images), and testing (294 images) subsets with an 8:1:1 ratio.The New-TunnelCrack dataset provides a diverse and high-quality benchmark under varying illumination and texture conditions, supporting the development and evaluation of real-time tunnel crack detection models.
提供机构：
Jiaxin Chen
