Automatic identification and characterization of fractures in borehole-surrounding rock based on borehole images

The internal structural planes of the surrounding rock are key factors controlling the strength and stability of rock masses, and quantitative analysis of their orientation of great engineering significance for tunnel construction safety and support design. To this end, we propose a method for automatically determining the orientation of structural planes based on borehole images. First, borehole camera technology is used to obtain two-dimensional cylindrically unfolded images of the borehole-surrounding rock walls, thereby constructing a dataset of fracture images in the surrounding rock; second, a segmentation network called ASA-UNet is developed to segment and identify fractures; based on the segmented fracture mask, an algorithm for decoupling fracture skeletons-which integrates topological connectivity and vector geometric criteria-is employed to automatically separate overlapping and intersecting fractures; Finally, an implicit spatial mapping method based on fracture pixel sequences is used to quantitatively characterizes the orientations of the primary weak structural planes, including the angle between the structural plane strike and the tunnel axis, as well as the dip angle. The results show that the ASA-UNet achieves a Mean Intersection over Union of 88.76%, mean Precision of 93.77%, mean Recall of 94.10%, and mean F1 of 94.03%, outperforming current mainstream semantic segmentation networks; the errors of the two calculated angles are within 5° and 10°, respectively, both of which fall within the engineering tolerance range.