Few-Shot Specific Emitter Identification: A Knowledge, Data, and Model-Driven Fusion Framework

Abstract

In the Industrial Internet of Things (IIoT) context, ensuring secure communication is essential. Specific Emitter Identification (SEI), which leverages subtle differences in radio frequency signals to identify distinct emitters, is key to enhancing communication security. However, traditional SEI methods often rely on large labeled datasets and complex signal processing techniques, which limit their practical applicability due to data acquisition challenges and inefficiency. To address these limitations, we propose a novel Few-shot Specific Emitter Identification (FS-SEI) approach named KDM. This method fuses deep learning with multi-modal data processing, utilizing a hybrid neural network architecture that combines handcrafted features, self-supervised learning, and few-shot learning techniques. Our framework improves learning efficiency and accuracy, especially in data-scarce scenarios. We evaluate KDM using open-source Wi-Fi and ADS-B datasets, and the results demonstrate that our method consistently outperforms existing state-of-the-art few-shot SEI approaches. For example, on the ADS-B dataset, KDM boosts accuracy from 60.99% to 75.34% as the sample count increases from 5-shot to 10-shot, surpassing other methods by over 10%. Similarly, on the Wi-Fi dataset, KDM achieves an impressive 88.94% accuracy in low-sample (5-shot) scenarios. The codes are available at https://github.com/tengmouren/KDM2SEI.