Pore type classification using multi-class classifiers: Application in rock physics modelling

Summary A key challenge to rock physicists is the quantification of pore geometry using accurate yet cost- and time-efficient methods. In the present study, pattern recognition techniques were used to formulate an intelligent algorithm for estimating pore geometry, with the methodology demonstrated on a carbonate oilfield as a case study. For this purpose, firstly, manual thin section analysis was performed by an expert geologist through polarized-light microscopy. Subsequently, thin section images were analysed using pattern recognition techniques, wherein different pore geometry features were estimated before image pre-processing and segmentation steps. Next, applying two different multi-class classifiers (support vector machine and k-nearest neighbours), pore types were obtained and classified according to the most widely used pore type classification scheme. The estimated pore geometry (in terms of pore type and aspect ratio) was subsequently incorporated into a rock physics model based on differential effective medium theory. Following with the research, P- and S-wave velocities were estimated considering matrix, pore type, and fluid properties. Finally, the results were interpreted and verified using measured well-logging data.