Generative Adversarial Networks for Exo-atmospheric Infrared Objects Discrimination

External atmospheric infrared (IR) target recognition is an important research topic of space surveillance systems. The different micromotion states of the target result in respective features in the obtained sequence of IR radiation intensities, and such differences are difficult to visually describe and difficult to extract efficiently. Due to the difficulty of spatial experiments, the target sample data is very limited. In the case of limited labeled samples, there are few methods that can effectively classify time series effectively, resulting in low classification accuracy and difficulty in meeting practical application requirements. We use generative adversarial networks(GAN) for semi-supervised learning, combining neural network classifiers with adversarial generation models for the classification of infrared grayscale time series of spatial targets. The effective classification of the target is achieved with a small number of labeled target samples. The experiment proves the effectiveness of the method and the classification effect is better than other methods.