Channel Attention Mechanism-based Multi-Feature Fusion Network for Specific Emitter Identification

In this paper, a channel attention mechanism-based multi-feature fusion network MF-RESNET is proposed for specific emitter identification (SEI). MF-RESNET takes the raw signal as input. Firstly, several traditional signal features, such as bispectrum, cycle spectrum and power spectrum, are achieved. Simultaneously, a residual neural network is utilized as the feature extraction network to extract the intrinsic characteristics from the raw signal. Then, the traditional signal features and the intrinsic features extracted by the residual neural network are fused based on the channel attention mechanism, where the potential radio frequency fingerprint (RFF) features are automatically exploited. Finally, the RFF features are sent to the simple CNN classifier network to output the probability of the category. Experimental results indicate that the proposed MF-RESNET can achieve 92.25% identification accuracy on automatic identification system (AIS) steady-state signal with 10 classes.