With the rapid development of the space industry, the importance of satellite health management technology has been increasingly recognized. Among the various subsystems in a satellite, the attitude control system is crucial for the reliable operation due to its highest failure rate. The data-driven fault diagnosis methods allow satellite faults to be detected and identified without building accurate mathematical models, so that proper adjustments could be made to fix faults. However, the challenge lies in the insufficient fault samples for training models. To address this problem, this paper proposes a fault diagnosis method for actuators in the satellite attitude control systems based on Res \(\_\) CBAM and Tradaboost. This method utilizes Res \(\_\) CBAM as the feature extractor to obtain semantic features of different samples. Then, the Tradaboost algorithm is employed to adjust sample weights for transfer learning, which improves the classification accuracy of the target satellite with insufficient training samples. The feasibility and efficacy have been confirmed through experiments.