The work aims to propose a multi-network joint infrared and visible image fusion algorithm, for the purpose of solving the problem of fusion image distortion and insufficient fusion of visible light image information in the fusion process of infrared and visible light images. Firstly, the inputted infrared and visible images were extracted by the encoder based on dense residual connection. Then, the fusion strategy was used to fuse the obtained feature map. Finally, the fused feature map was sent to the decoder based on GAN network. By optimizing the adversarial training with the visible light image, the fused image retained more details and background information of the visible image, and enhanced the visual effect of the image. Experiments show that, compared with the existing fusion algorithm, the proposed algorithm achieves better experimental results and has better expressiveness in subjective perception and objective evaluation.