A new state-of-the-art technique for textural and structural microimager data analysis using deep learning algorithms

Abstract

Borehole formation microimagers are a powerful tool for analyzing complex reservoirs, providing detailed information about the structural and textural features of formations. The development of state-of-the-art interpretation techniques can optimize existing approaches to microimager data analysis, enable the extraction of new object-level information, and significantly enhance the efficiency and quality of data interpretation. This study proposes a novel workflow for fullbore formation microimager data analysis based on processing a large and unique dataset using machine learning and computer vision techniques. The developed algorithms facilitate the automatic preprocessing of borehole microimager data and their automated structural and textural decomposition. The accuracy of object segmentation by convolutional deep neural networks exceeds 90%, while computer vision algorithms enable the analysis of the sizes, shapes, orientations, and topology of detected objects. The application areas of the proposed methodology include sedimentological analysis (thin-bed analysis); enhancement of core study workflow and formation tester evaluations (detailed characterization of reservoirs in intervals not covered by core samples); and advanced information for processing geological and geophysical data (reservoir modeling using deterministic approaches, distribution criteria for stochastic modeling and determining petrophysical parameters with high reliability).

Keywords

• borehole formation microimager
• convolution neural networks
• automated interpretation
• computer vision
• quantitative geological interpretation