Abstract: Establishing the correlation between the complex microscopic pore structure characteristics of the filter cake and the macroscopic seepage behavior is an important basis for solving the difficulty of coal slime dewatering. In order to characterize the pore structure of the filter cake accurately, four main minerals in the coal slime such as clean coal, quartz, kaolinite and montmorillonite were used as the research objects. The samples were scanned and imaged by CT, a 3D digital filter cake was constructed and a pore network model was extracted. Finally, the 3D display and quantitative characterization of the pore structure of the filter cake was realized. The classical K-C equation and the double fractal permeability model were deeply compared and analyzed in the filter cake permeability calculation. Based on fractal theory, Hagen-Poiseulle law and Darcy's law, combined with low-field nuclear magnetic resonance technology, bound water saturation and pore shape fractal dimension were introduced, and the existing fractal permeability model was revised to establish a filter cake microstructure. The permeability prediction model, the results show that the mineral composition in the slime filter cake is very complex, the filter cake formed by each mineral has obvious characteristic differences, the dewatering effect of clean coal and quartz is the best, and the pore size distribution of the clean coal filter cake is in the order of large pores. Mainly, but there are a certain amount of isolated pores inside, the connectivity is average, the pore tortuosity is the lowest, the porosity of the quartz filter cake is the largest, the connectivity is the highest, but the tortuosity is large. The filter cake formed by montmorillonite and kaolinite, the number of their pores is very small, and most of them are composed of pores below 10 μm, with large tortuosity, poor connectivity, which cause the poor dewatering performance. The filter cake of coal slime is mainly distributed in narrow strips, and the pore size is small. The overall porosity is low, the connectivity is poor, the tortuosity is high, and dehydration is difficult. The filter cake micro-permeability prediction model has high prediction accuracy for the permeability of quartz and clean coal filter cake, with relative errors of 1.34% and 1.15%, respectively. For the complex composition of slime filter cake and kaolinite filter cake, its permeability prediction error can be controlled within 5%, and the calculation error of the permeability of the montmorillonite filter cake can also be reduced to 13.42%.