Abstract: Most of the existing dehazing methods based on atmospheric scattering models still suffer from artifacts, color distortion, and incomplete dehazing in the face of complex and non-uniform haze, and a new GAN-based image dehazing algorithm D-GAN (Dehazing-GAN) is proposed to address the above problems. The generator of this network improved the network feature utilization by the global feature extraction subnet (GFES) and enhanced the reconstruction capability of the network for different scale details by using the multi-scale feature fusion subnet (MSFFS). Experiments show that the generative adversarial network model proposed in this paper has good robustness in non-uniform haze. Compared with FFA, SFNet, GCANet and other methods, it performs better on the objective evaluation indicators PSNR, SSIM, and has better performance in subjective evaluation.