The pore structure of marine sediments varies with the distribution of gas-hydrate, hence affecting the gas-water permeability. CT image is a conventional approach to view the internal structure, while for hydrate-bearing sediment investigation, rather poor resolution of obtained image has limited the accuracy of the analysis. Recently, super-resolution (SR) reconstruction techniques have been used to enhance the spatial resolution of CT images with varying degrees of improvement. Typical Image Pairs-Based SR (PSR) methods require higher resolution matching images for training, which is challenging for hydrate samples in dynamic temperature and pressure conditions. Here, we introduced a self-supervised learning (SLSR) method that only relies on a single input image to complete the process of training and reconstruction. We conducted a complete training to establish an end-to-end network consisting of two sub-networks, an SR network and a downscaling network. Self-built datasets from three hydrate samples with different sediment grains were trained and tested. Compared with the typical method, the SR results show that our method provides higher resolution while improving clarity. Moreover, in the subsequent calculation of porosity parameters, it has the highest consistency with the liquid saturation method. This study contributes to investigating the water seepage and energy transfer in the gas hydrate bearing sediments, which is particularly important for the exploration and development of marine natural gas hydrate resources. The image super-resolution method established by us has also a broad application prospect in the field of CT imaging.