A study on the automation of borehole image auto-recognition based on instance segmentation

To mitigate the time consumption and inefficiency of manual borehole image data recognition, this paper introduces a instance segmetation-based algorithm model called borehole image auto-recognition (BIAR). The algorithm initially employs the instance segmentation module of the you only look once version 8(YOLOv8)deep learning model to automatically detect and classify structures like dike filling and fractures in borehole wall images. Polynomial fitting formulas are then applied to the segmented mask coordinates to generate structural curves of the detected features. The dip and dip direction of these structures are calculated from the extrema of the fitted curves. Finally, the algorithm outputs the processed borehole images. A comparison between automatic and manual recognition using a dataset of actual borehole images demonstrates the algorithm's ability to quickly and accurately identify fractures and discontinuities, even in the presence of complex borehole wall information. The model's accuracy and robustness are also validated. Validation indicates that the BIAR algorithm can identify and map structural surface features in borehole images at a depth of 100 m within 30 s, achieving over 90% accuracy in identifying structural surfaces.This algorithm provides a highly efficient and accurate method for automatic borehole image analysis in engineering applications .