In comparison to visible light images, infrared images are less susceptible to factors like illumination and weather conditions, making them more suitable for applications in security, military, and various other fields. The utilization of infrared image generation techniques enables the rapid and cost-effective acquisition of a considerable number of simulated infrared images. Considering the characteristics of low contrast, low texture, weak information, and high noise in infrared images, this paper proposes a comprehensive strategy that includes a fusion of adaptive segmented linear transformation for image enhancement and arithmetic mean filtering for image denoising, in order to preprocessing infrared image data. Furthermore, an algorithm named Data Preprocessing-Arbitrary Style Transfer via Multi Adaptation Network (DP-MAST) is provided, based on a multi-adaptation network, to achieve the generation of multi-time infrared image data from a single time period to any multi-time periods. The experimental results show that compared to other image generation techniques, the similarity between the infrared images generated by DP-MAST and the real images has reached 91.4%.