BCGC-TunnelJoint: A High-Resolution Pixel-Level Dataset for Natural Tunnel Face Joint Segmentation under Blasting–Complex Geological Coupling Conditions

Overview The BCGC-TunnelJoint dataset contains high-resolution images of tunnel faces collected from drilling-blasting tunnel construction sites. The dataset focuses on the segmentation of natural geological joints under complex BCGC environments. Each image is manually annotated at the pixel level to identify joint regions. The dataset is specifically designed to support research on: Fine-grained tunnel joint segmentation Deep learning–based geological structure detection Robust segmentation under complex construction environments Intelligent monitoring of surrounding rock stability Data Characteristics The dataset captures several important characteristics of tunnel joints under BCGC conditions: Irregular and discontinuous joint morphology Thin and elongated structural features Severe class imbalance between joint and background pixels Environmental interference including dust, water stains, debris, and uneven illumination These characteristics make the dataset particularly suitable for evaluating segmentation algorithms in challenging engineering scenarios. Annotation Method All images are annotated using pixel-level semantic segmentation masks. The annotation process includes: Manual labeling by trained annotators Cross-validation by multiple experts Consistency checking to ensure labeling quality Each mask represents the binary segmentation of joint vs background. To ensure data privacy and prevent unintended information leakage, all images were processed to remove embedded EXIF metadata, including GPS coordinates, device information, and timestamps.