Robust Open-Set Signal Emitter Identification Using Collaborative Learning and Distributed Attention Networks

Signal emitter identification (SEI) has significant practical value in both military and civilian domains. In the military domain, it is used for target tracking and identification, while in civilian applications, it plays a crucial role in spectrum management and radio security. Current SEI technologies mainly focus on closed-set classification. However, in real-life scenarios, the electromagnetic environment is usually open and nonstatic. Existing algorithms suffer significant performance degradation when confronted with a large number of unknown class data. In the context of cyber–physical–social systems, SEI is required to break through the traditional single-node analysis model. Therefore, this article proposes an open-set recognition method based on collaborative learning of knowledge and data and distributed attention networks (CoDiNet). This method achieves robust recognition in complex electromagnetic environments through three-stage optimization involving collaborative learning of knowledge and data, distributed residual attention networks, and socialized dynamic decision-making. To validate the proposed methodology, empirical evaluations were performed on the ADS-B dataset, demonstrating that the CoDiNet framework achieves effective discrimination of both registered and emergent emitter categories under open-set conditions. The false alarm rate is significantly reduced, and the risk of misclassifying known classes as unknown classes is lowered. The method achieves a more balanced and stable performance, while also enhancing the robustness and reliability of the system. Overall, CoDiNet outperforms existing algorithms.