Multi-scale Specific Emitter Identification Via Self-attention-Based Feature Pyramid Network

Specific emitter identification (SEI) authenticates devices via identifying hardware characteristics. To improve identification accuracy and robustness, we design a novel neural network called a self-attention-based feature pyramid network (SA-FPN). The SA-FPN extracts multi-scale fingerprint features via the feature pyramid network (FPN) and makes predictions on the basis of those different scale features. Then, a squeeze and excitation network (SENet), embedded in SA-FPN, fuses predictions with adaptive weights and predicts emitter identities. Experimental results validate that the proposed method achieves higher accuracy and exhibits stronger robustness than other SEI methods.