Texture Segmentation of 3D X-Ray Micro Computed Tomography Images Using Supervised Methods

This research focuses on enhancing the petrophysical characterization of carbonate reservoir rocks using machine learning-based supervised textural classification methods. Carbonate reservoirs, known for their heterogeneity, present challenges in identifying textures like cemented porosity, neomorphism, and porous regions, which significantly influence properties such as porosity and permeability. Traditional methods often fall short in 3D analysis, prompting the use of Convolutional Neural Networks (CNNs) and specifically, the 3D U-Net architecture, renowned for its precise segmentation capabilities through multi-scale context and pattern recognition. This study employed 3D X-ray micro-computed tomography images of a rock sample, with the 3D U-Net model trained to classify the three textures. The segmentation process involved extracting feature maps and down-sampling through max pooling, followed by an up-sampling process to achieve accurate pixel-level classification. Training and validation were conducted using MATLAB, with the Adam optimizer facilitating efficient parameter tuning. The model's performance was evaluated through metrics such as pixel accuracy, intersection over union (IoU), and confusion matrices, demonstrating high accuracy and reliable classification. This approach highlights the potential of 3D CNNs and 3D U-Net in advancing the petrophysical analysis of complex carbonate reservoirs.