Abstract: The characterization of microstructures and prediction of transport properties in porous media serve as the fundamental theoretical basis for "dual-carbon" technologies such as energy storage and carbon capture, utilization and storage (CCUS). However, traditional methods face bottlenecks including insufficient statistical representativeness, high computational costs, and weak generalization across structures. Deep learning, with its powerful capabilities in nonlinear feature extraction, high-dimensional data processing, and integration of physical mechanisms, enables the construction of high-fidelity digital cores through super-resolution reconstruction, precise segmentation, and three-dimensional structure generation, establishing rapid and reliable mappings from microstructures to macroscopic heat and mass transfer properties. This review systematically summarizes the latest research advances in deep learning for digital structure characterization and transport property prediction of porous media, covering the technical evolution in digital structure characterization and model development in transport property prediction. It has been shown that the deep integration of data-driven approaches with physical constraints is significantly enhancing the efficiency and physical consistency of porous media research, promoting a paradigm shift from traditional experiment and simulation-driven approaches toward a more efficient and generalizable model-driven framework.