Rock Thin Section Image Search System Using Machine Learning Encoders
Mustafa A. Al Ibrahim, Robert J. Smith, Venkatesh Pathi, S. Hong · International Petroleum Technology Conference · 2025
Executive Summary Rock thin section analysis and interpretation are important processes in understanding formations. This study shows a new quantitative system to automatically search through thousands of legacy thin section images and their associated metadata and interpretation to find the closest match to an unlabeled sample. The system utilizes image-processing and neural-network-based encodings to build a database. Using this, it is possible to utilize existing legacy data as a guide to interpret new samples. The advantage of the system is that the associated metadata for the legacy thin sections does not need to be structured and images do not need to be labeled. Thus, we maximize the utilization of existing data without the need for these time-consuming tasks that are usually needed for most formulations of machine learning problems.