Detection of rock fragmentation from borehole imaging logs using a deep learning method

ABSTRACT In underground engineering, rock fragmentation is a critical factor affecting construction safety and stability, and its efficient and accurate delineation is essential for subsequent engineering activities. Conventional methods, which rely mainly on manual measurements, are inefficient, labor-intensive, and prone to errors. To overcome these limitations, a deep learning-based approach is proposed to automatically identify rock fragmentation zones and calculate related parameters. A diverse dataset of borehole images across various lithologies was compiled to improve model robustness, and multiple models with different training parameters were evaluated to determine the optimal configuration. The selected model was then applied to detect fragmentation zones and extract relevant parameters, with its effectiveness verified in real engineering scenarios. The results demonstrate that the You Only Look Once-based model achieves high predictive accuracy (85%) and reliably identifies fragmentation zones, showing strong consistency with borehole wave velocity data, thereby confirming its precision and applicability. This method enables rapid, accurate, and automated characterization of rock fragmentation zones, substantially improving efficiency and offering a novel deep learning application in geoscience.