Lithology Identification in Carbonate Thin Section Images of the Brazilian Pre-Salt Reservoirs

Understanding the composition of carbonate sedimentary rocks is crucial in assessing their potential within presalt oil reservoirs and its several areas like: exploration, reservoir characterization and production. This study delves into the significance of these rocks within the oil and gas industry, emphasizing their mineral composition, diagenetic alterations, and their impact on reservoir analysis. With a focus on automated analyses using petrographic data, this work addresses the challenges posed by unbalanced classes and heterogeneous rock structures. Leveraging a diverse methodology involving image preprocessing, data augmentation, and advanced machine learning and image processing techniques like ViT (Vision Transformer), MixUp Data Augmentation, and Focal Loss, the study attains good accuracy results in lithology classification. Notably, RandAugmentation and MixUp prove pivotal in enhancing model performance. The findings underscore the potential of employing deep learning and computer vision techniques to automate lithology classification in Brazilian pre-salt carbonate thin sections, suggesting avenues for future improvements such as database expansion and increased class variety.