DOI: A Systematic Framework for Incremental Identification of Specific Emitters in Open-Set Scenarios
Fei Teng, Wenqiang Shi, Yingke Lei, Hu Jin, Bisheng Pan, Yourui Wang · IEEE Transactions on Cognitive Communications and Networking · 2025
Specific emitter identification (SEI) technology can be used for identity recognition, passive authentication, and safety management of target emitters, which is of great significance in enhancing information security. In response to the demand for identifying the continuously increasing unknown emitters in real dynamic scenarios, this paper proposes a novel and comprehensive incremental identification framework, DOI, for unknown emitters in open-set scenarios. DOI accomplishes the tasks of detecting, classifying, and incrementally recognizing unknown emitters. Specifically, in DOI, we firstly introduce a data preprocessing method and a lightweight network architecture, which improve inter-class discrimination and compactness within classes while controlling computational costs. Secondly, we design an open-set recognizer that classifies unknown classes while detecting them. Finally, we employ a recall strategy to select more representative known classes. Based on this, incremental learning (IL) is applied to unknown classes. DOI’s performance on self-collected and publicly available datasets demonstrates the superior capabilities of the data preprocessing method and open-set recognizer, as they can achieve a very high accuracy in detecting and classifying unknown classes. Furthermore, in the incremental task, DOI achieves the final accuracy rates of 91.25% and 90.84% when adding 5 and 4 unknown classes in each incremental session, respectively, outperforming recent state-of-the-art algorithms. At the same time, we further demonstrate the effectiveness and rationality of DOI through ablation experiment.