The verification of the adaptability and robustness of infrared target recognition algorithms in complex and variable environments requires a large amount of target and interference test image data. Currently, various experimental verification methods generally suffer from insufficient scenario coverage and inadequate realism. To address this, this paper proposes a method for generating complex infrared interference scenarios based on few-shot data. The method first constructs a deep learning algorithm based on a semantic generation model to extract three-dimensional modeling features of targets and interference. Then, it combines knowledge transfer and deep diffusion generative models to achieve controllable simulation generation of typical infrared interference scenarios based on few-shot data. Finally, it uses image fusion algorithms to fuse infrared target and interference scenario images, achieving a close fit to the sample data. Experimental results have shown that the method in this paper has an effective ability to generate complex infrared interference scenarios, and the generated results have high credibility.