Abstract: Ground Penetrating Radar inversion algorithms have unique advantages in the fine characterization of geological bodies. Taking digital core modeling as an example, this paper proposes a core reconstruction method that combines the quartet structure generation set (QSGS) and conditional generative adversarial network (CGAN) in response to the problems of high cost, low reuse rate, and poor reconstruction quality of traditional digital core reconstruction techniques. Firstly, QSGS is used to obtain 3D four phase core images containing pore phase, quartz phase, calcite phase, and skeleton based on pore and mineral parameters, and then input them as conditional images into the CGAN network, produce multicomponent 3D grey scale core images. The experimental results indicate that when CGAN is trained on four phase core images containing both pores and minerals, the trained model can not only learn the characteristics of the real pore structure in the core, but also capture the morphology, distribution, and grey scale representation of minerals. Meanwhile, the reconstructed 3D core meets the volume requirements of pore space, quartzite and calcite, and can represent various components in the real grayscale core image to a large extent accurately, which is expected to provide a more accurate core modeling support for geologic analysis based on Ground Penetrating Radar inversion.