Infrared and visible image fusion of generative adversarial network based on multi-channel encoding and decoding

•The structure of the triple encoding decoder is designed for feature extraction. Infrared and visible separate paths provide more detailed contrast and contour texture details.•A contrast enhancement module is designed after the infrared path structure of the encoder in the generator to enhance the contrast information of the target region.•Double discriminators are used to discriminate the generated images. In order to further improve the feature learning ability of the generator, an incentive module is designed at the end of the discriminator.•Specific loss functions are designed. Specifically, gradient loss and contrast loss are designed in the generator loss function, and the learning and training process is more accurate by changing the concrete function.•A large number of experiments have proved that the proposed algorithm can achieve significant fusion results with visual effects, and because the trained model is directly utilized in the test stage, the real-time performance of the algorithm is also very high. Image fusion can overcome the shortcomings of unclear, inaccurate and incomplete target description of single sensor image, achieve more accurate expression of target, and lay a foundation for subsequent image analysis and processing. Aiming at the existing algorithms based on generative adversarial networks only relying on the adversarial training of discriminators to achieve image fusion, and the balance of the respective features of infrared and visible images is not comprehensive or even fine enough, an infrared and visible image fusion algorithm based on multi-channel encoding and decoding generative adversarial network was proposed in this paper. Compared with the existing image decomposition methods, a triple codec structure is designed for feature extraction, which can use the generator itself to get a fusion image with extraordinary effect, and then use the discriminator to improve the quality of the fusion result. Secondly, a contrast enhancement module is designed after the infrared path structure of the encoder in the generator to enhance the contrast information of the target area. In addition, a dual discriminator model is designed, and in order to improve the feature learning ability of the generator, an incentive module is designed at the end of the discriminator. Finally, gradient loss and contrast loss are designed, and the training speed is improved by combining specific functions. Experimental results show that compared wi