The structural characteristics of nanopores are known to significantly affect the wetting effect in coal seam water injection. Currently, the three-dimensional characterization of nanopores in coal relies mainly on digital images, whereas poor image resolution and segmentation methods pose significant challenges. Therefore, using coal samples from Wudong Coal Mine in China as an example, cryo-focused ion beam scanning electron microscopy (cryo-FIB-SEM) and deep learning segmentation methods were implemented to accurately characterize the nanopores and water distribution. In the obtained pore structure, the number of isolated pores was higher than that of connected pores, while the volume of connected pores was significantly larger than that of isolated pores, comprising the key path and storage space for external water to enter the coal body. The water content of isolated pores mainly depends on the permeability of the coal matrix. The connectivity of single pores can be characterized by the coordination number, whose increase leads to the number of pores exponentially decreasing. The connectivity of pore clusters depends on the number of internal branches. The number of branches in the pore cluster increases exponentially with the increasing total length, total volume and average radius of the cluster, and the connectivity is correspondingly enhanced. The increase in pore size enhances the shape factor, surface area and connectivity of pores while reducing tortuosity, which in turn facilitates coal wetting. The accurate characterization of coal nanopores in this study helps to scientifically evaluate the effect of coal seam water injection, highlighting the importance of increased pore size and improved pore connectivity for enhanced water injection effectiveness.