Space target classification is regarded as a key technology for space situational awareness and security assurance. However, the generalization performance of conventional deep learning models remains limited under the small-sample regime due to scarce high-quality measured data, imbalanced class distribution, and environmental interference.To overcome this bottleneck, a hybrid method combining transfer learning (TL) and generative adversarial networks (GANs) was proposed. In this method, pretrained parameters and feature priors were transferred to the GAN to improve the quality of synthetic samples. The sample set expanded by the GAN was used to optimize the fine-tuning process of the transfer-learning model and enhance the adaptability to the target domain.The method was applied to the classification of small-sample space-target spectral data. The spectral data acquired by the 1.2 m photoelectric telescope at Changchun Station were processed by the proposed method. The similarities between the generated data and the original data are 81.97%, 82.06%, and 84.85% for the sample-based, feature-based, and sample-feature-fusion TL-GAN methods, respectively. The overall classification accuracies are 78.32%, 78.41%, and 81.05%, respectively. Compared to the case without generated data, the classification accuracies are increased by 2.25%, 2.28%, and 2.08% respectively.The proposed method improves feature-learning efficiency and classification accuracy. A new technical route for small-sample space target classification is provided by this method, offering both theoretical and practical support for the engineering application of space situational awareness systems.