In aeronautical communications, because of that data are imbalance, and being lack of minority class signal samples, there is a drop in the classifier performance in modulation signal classification tasks under complex electromagnetic environments, this paper proposes a classification method for unbalanced modulation signals based on deterministic oversampling. The method synthesizes minority class signal samples to balance the dataset and reduce the impact of data imbalance on classification performance. Based on the method at the RadioML 2016.10a dataset, 11 modulation types are selected under 5 signal-to-noise ratios (-8 dB, -4 dB, 0 dB, 4 dB, 8 dB) and 4 imbalance scenarios are constructed. The experimental results show that compared to the imbalanced dataset, the proposed method improves classification accuracy by 2.78%, 0.92% and 3.45% on MsmcNet, ResNet50, and DenseNet121 network models respectively, and compared to the traditional SMOTE method, the proposed method demonstrates still better performance in handling multi-class imbalance problems. And this method is enabled to effectively improve the accuracy in modulation signal classification in complex aeronautical communication environments, especially under complex electric-magnetic environments.