EUVIMEDEUVIMED entdecken
Crossref

Contrastive Incremental Learning With Relational Distillation and Selective Replay for Specific Emitter Identification

Dongming Wu, Junpeng Shi, Zhihui Li, Mingyang Du, Kui Zhang, Fangling Zeng

IEEE Transactions on Aerospace and Electronic Systems · 2026 · 62 · 19158-19170

DOI: 10.1109/taes.2026.3742296 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2026. Verfasst von Dongming Wu und 5 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Specific emitter identification relies on extracting subtle hardware fingerprints to authenticate devices. Unlike conventional methods based on static datasets, practical electromagnetic environments are dynamic, requiring systems to incrementally learn unseen emitters without the need for retraining from scratch. Nevertheless, contemporary deep learning paradigms are prone to catastrophic forgetting and often fail to preserve feature discriminability amidst time-varying channel conditions. To address these challenges, we propose a contrastive incremental learning with relational distillation and selective replay. Specifically, we replace the conventional cross-entropy loss with supervised contrastive learning. This approach pulls signals from the same emitter together on a hyperspherical embedding space, significantly enhancing noise robustness and intraclass compactness. Subsequently, we introduce a relational distillation mechanism that aligns the instance-to-instance similarity matrices of the current and past models to mitigate catastrophic forgetting. Distinct from traditional output-logit distillation, this method explicitly preserves the structural topology of the learned feature space. Finally, we design a selective replay strategy based on a contrastive contribution metric, which retains high-value exemplars that best represent the class distribution rather than random sampling. Extensive experiments on the real-world dataset demonstrate that our method outperforms existing incremental learning baselines in recognition accuracy.

Zitieren

APA

Wu, D., Shi, J., Li, Z., Du, M., Zhang, K., & Zeng, F. (2026). Contrastive Incremental Learning With Relational Distillation and Selective Replay for Specific Emitter Identification. IEEE Transactions on Aerospace and Electronic Systems, 62, 19158-19170. https://doi.org/10.1109/taes.2026.3742296

Vancouver

Wu D, Shi J, Li Z, Du M, Zhang K, Zeng F. Contrastive Incremental Learning With Relational Distillation and Selective Replay for Specific Emitter Identification. IEEE Transactions on Aerospace and Electronic Systems. 2026;62:19158-19170. doi:10.1109/taes.2026.3742296

BibTeX
@article{Wu2026,
  author = {Wu, Dongming and Shi, Junpeng and Li, Zhihui and Du, Mingyang and Zhang, Kui and Zeng, Fangling},
  title = {Contrastive Incremental Learning With Relational Distillation and Selective Replay for Specific Emitter Identification},
  journal = {IEEE Transactions on Aerospace and Electronic Systems},
  year = {2026},
  volume = {62},
  pages = {19158--19170},
  doi = {10.1109/taes.2026.3742296}
}
Kostenlos starten →