Generative adversarial network (GAN) architecture is employed to tackle the inverse scattering problem of buried dielectric objects in half-space. Traditional iterative methods aimed at resolving the inverse scattering problem of buried dielectric objects have encountered a variety of difficulties, such as highly nonlinear phenomenon, high computational costs for half-space Green’s function, and missing measured scattered field information at the lower half of the object. The generator of GAN learns to generate more realistic images, while the discriminator of GAN improves its ability to identify fake images through a game-like process. The iterative process stops when the image generated by the generator is indistinguishable from the real image. In addition, we also analyze and compare the reconstruction outcomes obtained using both GAN and U-Net. Numerical outcomes show that GAN can efficiently reconstruct images with higher reliability than U-Net for buried objects with different dielectric permittivities and handwritten shapes. In summary, our proposed method has opened up a new avenue for imaging buried objects by adopting a deep learning network technique.