Abstract: ‍ ‍The rapid development of drone technology in recent years has played an important role in military reconnaissance， cargo transportation， geographic surveying， and mapping， and in civilian fields such as agricultural monitoring， natural-disaster assessment， and aerial photography. The versatility and accessibility of drones have revolutionized numerous industries， offering new perspectives and capabilities that were previously unattainable. However， alongside these advancements， the widespread use of drones has introduced significant safety and privacy concerns， particularly in drones employed for unauthorized surveillance or in restricted airspaces. Consequently， the precise identification of drones has become critical for both national and civil aviation safety. Recognizing these challenges， researchers and technologists have adopted deep learning to address the aforementioned challenges. Deep learning， a machine-learning method characterized by the use of algorithms inspired by the structure and functions of the brain called artificial neural networks， has shown exceptional promise in pattern recognition and anomaly detection. In the context of drone technology， a deep learning-based method for the specific emitter identification of a drone has emerged as a cutting-edge approach. This method fundamentally relies on analyzing the radio frequency （RF） signals emitted by drones. These signals are processed using advanced deep-learning models， leading to the accurate identification and classification of different drone models and types based on their unique signal characteristics. However， implementing this method faces challenges. First is the requirement of substantial training data that are independent and identically distributed. In real-world scenarios， the collection and annotation of drone RF data are fraught with difficulties， including limited data availability and inconsistent data quality. Moreover， there is often a significant discrepancy between the distribution of training data and that of test data， which can severely impact the ability to generalize a model for new or unseen conditions. To address these issues， this paper proposes a method for the individual identification of drone radiation sources， based on domain adaptation models. Domain adaptation， which is a technique in machine learning， aims to enable a model trained in one domain （the source domain） to perform effectively in another （the target domain）. The proposed method employs several innovative techniques， including resampling with a bootstrap method. Resampling the drone dataset using the bootstrap method increases the diversity of the samples within the dataset. This approach helps mitigate the problems associated with limited and imbalanced data. Additionally， transformer encoders are combined with domain adversarial neural networks （DANNs）. The integration of transformer encoders， which are known for their effectiveness in processing sequential data， with DANNs， enables the model to optimize feature representation under a Gaussian distribution. This fusion enhances the model’s adaptability to new and varied environments. The maximum mean discrepancy （MMD） is also utilized. Incorporating the MMD as a metric reduces the distributional discrepancy between the source and target domains during training. This is pivotal in enhancing the model’s generalization capabilities. A weighted voting mechanism is employed to aggregate the outputs of various models. This technique boosts the generalization power of the overall system and improves its adaptability and accuracy under diverse environmental conditions. Empirical results demonstrated the efficacy of this method. The proposed method was tested using three typical environments and six domain adaptation scenarios， and its recognition rate was approximately 5% better than those of the existing techniques. This performance indicated a successful balance between the accuracy， generalization， and reliability demands of individual drone radiation source identification. In conclusion， the proposed method not only addresses the pressing need for enhanced drone safety monitoring but also sets a precedent for the application of deep learning in specialized data scenarios. The blend of advanced machine learning techniques with domain-specific knowledge paves the way for more robust and adaptable systems that can tackle the complex challenges posed by the rapid evolution of drone technology.