Lithology classification of whole core CT scans using convolutional neural networks

Abstract X-ray computerized tomography (CT) images as digital representations of whole cores can provide valuable information on the composition and internal structure extracted from wells. Incorporation millimeter-scale core CT data into lithology classification workflows result in high-resolution description. In this study, we use 2D scan image slices to train a convolutional neural network (CNN) whose purpose is automatically predict well Norwegian continental shelf. The are preprocessed prior training, i.e., undesired artefacts flagged removed further analysis. training include expert-derived lithofacies classes obtained by manual trained classifier used set test that unseen classifier. prediction results reveal distinct predicted with high recall (up 92%). However, there misclassification rates associated similarities gray-scale values transport properties. To postprocess acquired results, identified merged similar through ad hoc analysis considering degree confusion matrix aided porosity–permeability cross-plot relationships. Based analysis, four rock classes. Another CNN resulting generalize well, higher pixel-wise precision when detecting thin layers bed boundaries compared Thus, provides additional complementing already existing type Article Highlights A workflow for automatic using introduced. proposed shows lithology-dependent accuracies. exploited tool identify properties generate hierarchies.