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

In underground engineering, rock fragmentation is a critical factor affecting construction safety and stability. The rapid and accurate delineation of fragmented rock formations is of significant research importance for subsequent engineering projects. Current methods for characterizing rock fragmentation zones primarily rely on manual measurements, which are time-consuming, labor-intensive, and prone to inaccuracies. To address this, a deep learning-based method is proposed to identify rock fragmentation and calculate related parameters. First, to enhance the model’s robustness, a large collection of borehole images with varying lithologies was gathered, forming a comprehensive training dataset. Second, models with different training parameters were evaluated and compared to identify the most suitable model for further analysis. The selected model was then applied to identify rock fragmentation zones and compute relevant parameters. Finally, the model’s effectiveness and feasibility were validated through its application in real engineering scenarios. Test results show that the YOLO-based model demonstrates excellent predictive performance, achieving an accuracy of 85%. Furthermore, the model performs well in identifying rock fragmentation zones, showing consistency with real borehole wave velocity data, thus validating its precision and feasibility. This method enables rapid and accurate identification of fragmented rock zones, significantly improving efficiency. As an innovative application of deep learning in geoscience, it offers a novel approach for the rapid and accurate characterization of rock fragmentation zones.