Carbon capture and storage is an essential technique for large-scale, low-carbon fossil fuel mitigation and effectively reduces the release of CO2 into the atmosphere. This gas storage is beneficial to the environment, as it prevents air pollution and prevents contamination of local fauna and flora. The injected CO2 serves to reduce oil thickness and improve the recovery productivity of the remaining unproduced oil. Deep learning frameworks using convolutional neural networks have introduced fast and robust methods for automated image processing. This work proposes a method for automatic segmentation of rock porosity in computed tomography images. It uses preprocessing to reduce the processing required on the image, employing an architecture that, despite the increased parameters, does not significantly affect its speed. It also uses postprocessing to aid in the extraction of rock information, such as connected porosities and possible fluid pathways. This approach achieved 99.15% F1-Score, 99.28% sensitivity, and 99.00% accuracy. Furthermore, these porosities are analyzed and the flows are visualized.