Aiming at the problems of poor contrast and low signal-to-noise ratio of traditional ghost imaging, a fusion ghost imaging method based on frequency domain decomposition is proposed. This method transforms the speckle pattern obtained in the reference path to the frequency domain, selected an appropriate threshold to decompose it into high-frequency speckle and low-frequency speckle, and performs correlation calculation between the high-frequency speckle and the value obtained by the bucket detector to get high and low ghost image, and finally the inverse non-subsampled shearlet transtransform is used to reconstruct the final ghost image. By using PSNR and contrast as the evaluation index, the effectiveness of the fusion ghost imaging method was verified by 4 sets of experimental simulations. The simulation experiment results show that the PSNR/contrast of the fusion ghost imaging is improved by 41%/173% and 27%/135% on average compared with the computational ghost imaging and differential ghost imaging methods.