Image Segmentation of 3D X-ray Micro-computed Tomography Images

The objective of this research is to explore the effectiveness of different deep learning models in accurately segmenting 3D X-ray Micro Computed Tomography images of carbonate rocks. Four models with different architectures and training parameters were evaluated, and the results suggest that models using a smaller patch size demonstrated superior segmentation performance, regardless of the architecture used. The models were trained and validated using cropped regions of interest extracted from the original dataset. This study emphasizes the importance of patch size selection in influencing segmentation outcomes and highlights the robustness of the developed models in accurately segmenting rock structures. These findings provide valuable insights into model selection and parameter optimization for enhanced segmentation accuracy, which can contribute to the advancement of image segmentation techniques for rock analysis.