Logging-while-drilling (LWD) borehole images are very important data to support formation characterization and drilling operations. The manual interpretation of this data is a time-consuming task, limited by inconsistencies and uncertainties. We propose a deep-learning (DL)- based supervised method to automatically correlate geological features in low-resolution LWD image logs.
Additionally, we tested two learning strategies, namely standard learning (SL) and curriculum learning (CL), to critically analyze the differences in the application on both synthetic and field data. Our results show that these DL models can effectively replace manual labor in dip picking but highlight the need for human intervention to validate and classify the correlated features, proving the utility of the semi-automatic paradigm.
