PHASE SEGMENTATION OPTIMIZATION OF MICRO X-RAY COMPUTED TOMOGRAPHY RESERVOIR ROCK IMAGES USING MACHINE LEARNING TECHNIQUES

Abstract

We studied the performance and accuracy of some basic segmentation techniques in the analysis of the pore space and matrix voxels obtained from a 3D volume of X-ray tomographic (XCT) grayscale rock images. The segmentation and classification accuracy of unsupervised (K-means, modified Fuzzy c-means, Minimum cross-entropy, and Type-2 fuzzy entropy) and supervised Naïve Bayes methods were tested using an XCT tomogram of a carbonate reservoir rock. K-fold- cross-validation techniques were applied in the evaluation of the accuracy of the unsupervised and supervised machine learning classifiers. The average porosity obtained was 31 ±6%, in good agreement with the ground truth image obtained by manual segmentation. In general, the accuracy of segmentation results can be strongly affected by the feature vector selection scheme, since it is difficult to isolate a particular machine learning algorithm for the complex phase segmentation problem. Therefore, our study provides a segmentation scheme that can help in selecting the appropriate machine learning techniques for phase segmentation.