Specific Emitter Identification Method based on Deep Ensemble Learning

In this paper, two deep ensemble algorithms, the hierarchical and multi-stage feature fusion(HMSFF) and Cooperative decision(Co_decision), are designed for specific emitter identification(SEI). Extracting the differential constellation trajectory amplitude, the energy spectrum as well as the spectrum symmetry as radio frequency fingerprint features, the proposed algorithms exhibit better recognition accuracy and robustness to time variation than existing researches. Furthermore, in the case of small number of training samples two data augmentation methods, noise augmentation and multipath channel augmentation, are used to further improve the recognition performance.