Leveraging Regeneration Learning for Improved Acoustic Image Log Quality Assessment

Summary Regeneration learning is a novel machine learning paradigm designed to effectively handle different data modalities with unbalanced information content. The recent BERGSOM framework leverages this approach to regenerate acoustic image logs from conventional logs. To ensure the quality of these ML-generated images, effective assessment metrics are essential. However, traditional image quality assessment (IQA) metrics often prove inconclusive when evaluating acoustic borehole image logs generated by models such as BERGSOM. This study investigates the effectiveness of the BERGSOM framework using a real-world dataset of high-resolution acoustic image logs and conventional logs from five wells drilled in a heterogeneous Brazilian pre-salt carbonate reservoir. The results indicate that the intrinsic BERGSOM quality metrics significantly outperform conventional IQA metrics when assessing the quality of acoustic image logs. This finding highlights the critical role of the BERGSOM methodology in improving petrophysical data analysis and supporting well drilling and completion decision making.