In order to characterize the microstructure of pellets,a 3D reconstruction method based on industrial CT is proposed to address the shortcomings of traditional two-dimensional images in terms of structural integrity and spatial correlation. In the image processing stage, the slices are preprocessed by using joint filtering, adaptive histogram equalization and super-resolution model to improve the grayscale distribution and phase boundary separability. The method retains the spatial distribution characteristics of pores, liquid phases and hematite phases through voxel construction. Combined with the annotation mechanism based on volume distribution and the gray wolf optimization algorithm,the three-phase threshold automatic optimization is completed (the optimization results are about 80. 03 and 102. 25). The three-dimensional phase statistics show that the volume fractions of hematite,liquid phase and pore are stable at 82% ~ 86%,8% ~ 11% and 4% ~ 6%,respectively,which reduces the statistical fluctuations caused by twodimensional slicing. On this basis,the fractal dimension of the pellets is calculated and compared with the two-dimensional weighted results,and the results show that the two-dimensional dimensions are mainly distributed in 1. 2 ~ 1. 6,while the three-dimensional dimensions are greater than 2. 5,and the difference between different directions is as low as 0. 007, showing good isotropy and structural stability. The three-dimensional method overcomes the direction dependence and truncation error,improves the objectivity of volume distribution and structural complexity indexes,and provides basic data support for pellets structure characterization and metallurgical process analysis.