Semi-Supervised Specific Emitter Identification Method Using Metric-Adversarial Training

Specific emitter identification (SEI) plays an increasingly crucial and potential role in both military civilian scenarios. It refers to a process discriminate individual emitters from each other by analyzing extracted characteristics given radio signals. Deep learning (DL) deep neural networks (DNNs) can learn the hidden features of data build classifier automatically for decision making, which have been widely used SEI research. Considering insufficiently labeled training samples large-unlabeled samples, semi-supervised learning-based (SS-SEI) methods proposed. However, there are few SS-SEI focusing on extracting discriminative generalized semantic In this article, we propose method using metric-adversarial (MAT). Specifically, pseudo labels innovatively introduced into metric enable (SSML), objective function alternatively regularized SSML virtual adversarial (VAT) is designed extract The proposed MAT-based evaluated open-source large-scale real-world automatic-dependent surveillance–broadcast (ADS-B) set Wi-Fi compared with state-of-the-art methods. simulation results show that achieves better performance than existing when ratio number all 10%, accuracy 84.80% under ADS-B 80.70% set. Our code be downloaded https://github.com/lovelymimola/MAT-based-SS-SEI .