Super-Resolution Imaging of Multiphase Fluid Distributions in Porous Media Using Deep Learning

Abstract

Super-resolution imaging techniques use deep learning to create large-scale, high-resolution images by combining a low-resolution image encompassing a large volume with high-resolution images on a smaller volume; however, applications to date have been limited to determining the pore structure only. We have successfully applied an enhanced deep super-resolution (EDSR) method to three-dimensional X-ray images of two fluid phases in the pore space of water-wet and mixed-wet Bentheimer sandstone, producing high-resolution results that capture both the pore space and two fluid phases within it, while expanding the field of view. We calculated and compared the geometrical and physical properties, including porosity, permeability, saturation, interfacial area, interfacial curvature, and contact angle derived from high-resolution, super-resolution, and low-resolution images. This comparison confirms that our super-resolution outcomes are consistent with the ground truth and far superior to low-resolution results.