Research on Infrared Scene Generation Method Based on Asymmetric Small Sample Images

To address the issues of high contrast between the output target and the scene, excessive differences from the real scene, inaccurate representation of the weather factor, the need for a large amount of infrared data to train the network and incomplete infrared simulation data sets in the existing infrared simulation techniques, an asymmetrical few-shot infrared image generative network (AFIGN) is proposed. AFIGN adopts the CycleGAN architecture as the main architecture and designs a small sample generation network module as the generator, which can solve the problem that symmetric data sets are needed in the training process and the real infrared scene data are small to meet. The designed network can obtain infrared scene data with high similarity even under the condition of a 10:1 asymmetric small sample data set, which shows that AFIGN can improve the simulation accuracy and efficiency of existing infrared semi-physical simulation system under the condition of limited infrared simulation data set.