A comprehensive review of 3D digital mapping-based automated rock discontinuity detection: emerging technologies, challenges and future prospects

Accurate detection and characterization of rock discontinuities are essential for rock slope stability assessment, tunnel excavation, rockfall hazard evaluation and rock mass classification. Conventional field methods, including geological compass measurements, scanline surveys and manual mapping, remain important but are limited by restricted spatial coverage, operator subjectivity, time consumption and safety risks in steep or inaccessible rock masses. This review critically examines recent progress in automated and semi-automated rock discontinuity detection using 3D digital mapping, advanced remote sensing and artificial intelligence-based methods. The reviewed detection approaches include LiDAR-based terrestrial and aerial laser scanning, UAV photogrammetry, hyperspectral imaging, 3D point-cloud processing and classical image-processing techniques such as edge detection. Particular emphasis is placed on AI-driven discontinuity detection, including convolutional neural networks for 2D image-based recognition, semantic segmentation models such as U-Net and DeepLabV3+, object and instance segmentation frameworks such as YOLO, Faster R-CNN, Mask R-CNN, Mask R-CNN–TD and YOLACT++, and advanced architectures including 3D point-cloud networks, transformer-based models, graph neural networks, transfer learning and hybrid 2D–3D fusion models such as CNN–PointNet. This review shows that LiDAR- and UAV-derived point clouds can provide reliable discontinuity orientation, spacing, joint-set and block-geometry information, supporting stereonet-based kinematic analysis, rock mass quality assessment, RQD estimation and identification of potentially unstable blocks. Despite these advances, major challenges remain, including limited benchmark datasets, weak model generalization across heterogeneous geological conditions, occlusion, vegetation interference, variable lighting, point-density effects, computational cost, insufficient ground validation and limited integration with practical geotechnical design workflows. Future research should prioritize standardized open datasets, uncertainty-aware and explainable AI, cross-site validation, domain adaptation, robust multimodal sensing–AI fusion, and direct integration of automated discontinuity outputs with RMR, SMR, Q-system, GSI, DFN modelling, and slope stability analysis.