The invention discloses a space target classification method based on a multilayer adversarial network, and the method comprises the following steps: 1, carrying out the sequential multi-frame projection of an imaging target through an imaging load, and obtaining a space image; 2, performing feature augmentation on the space image to obtain an augmented space image; 3, based on the top-layer adversarial network, performing foreground extraction on the augmented space image, and extracting foreground information; and step 4, based on conditional convolution adversarial training, performing progressive two-stage adversarial on the foreground information, extracting a general representation vector, and finishing space target classification. According to the invention, the classification problems that various types of targets exist in an imaging load acquisition image under a complex state and relative motion exists can be solved; reliable feature extraction can be achieved through gamingbased on the adversarial network, accurate classification of various complex targets is achieved, targeted application is developed, the on-orbit autonomous operation capacity of the active spacecraftis effectively improved, and important guarantee is provided for autonomous navigation and safe survival of the spacecraft.