Zufar Kayumov

Petrographic analysis of thin sections of terrigenous rocks plays a key role in characterizing reservoir properties, enabling the assessment of textural and structural features as well as porosity—essential for reconstructing depositional environments and post-sedimentary alterations. Conventional manual analysis of thin sections is time-consuming and subjective, necessitating its automation. This study proposes an automated method for analyzing thin-section areas of quartz sandstone with carbonate cement based on deep learning, combining instance segmentation using the Mask R-CNN model with semantic segmentation employing SemanticFPN. This ensemble approach achieves high segmentation accuracy, as confirmed by Pixel Accuracy and Intersection over Union metrics, significantly accelerating petrographic data processing while enhancing objectivity and reproducibility. The method holds practical value for geological and petroleum exploration studies.