Cascade Characterization of Rock Fractures with Unsupervised Image Semantic Segmentation, Enhancement and Virtual Mapping

Intelligent identification of fractures and the subsequent utilization of results for attribute characterization are essential in the structural analysis of tunnel rock. A systematic cascade method for characterizing rock fractures was proposed, which integrates unsupervised image semantic segmentation, enhancement, and virtual mapping. It leverages the complementary strengths of the DexiNed model for unsupervised edge detection, Gabor filters for trace enhancement, and RBI-based evaluation metrics for structural characterization. The DexiNed deep learning algorithm was applied for edge detection to perform unsupervised fracture identification. Gabor filtering was introduced to enhance the fracture traces, which were subsequently segmented and analyzed using the DBSCAN clustering algorithm. The extracted fracture attributes were then characterized through a combination of Risk-Based Inspection (RBI) evaluation and virtual mapping. The proposed method has been applied in a tunnel project in Xinjiang, China and demonstrates high accuracy and practical feasibility. Research findings showed that the method delivers comprehensive structural data, providing a robust solution for mining and characterizing fracture attributes in tunnel engineering.