Radar Specific Emitter Identification Based on AE-VMD and ECAResNet

Radar specific emitter identification is one of the core technologies in electronic support measures and battlefield situational awareness. Existing radar specific emitter identification methods based on Hilbert-Huang Transform （HHT） and deep learning are limited by poor selection of decomposition parameters and low identification accuracy. To address these issues， a radar specific emitter identification method based on variational mode decomposition （VMD） parameter optimization and Efficient Channel Attention-Residual Neural Network （ECAResNet） was proposed， combining intelligent optimization algorithms with signal processing. First， the signal was decomposed into multiple optimal modal components by means of the Alpha evolution （AE） optimization algorithm， combined with VMD to achieve an adaptive optimal decomposition of the parameters； second， the Hilbert transform was applied to the decomposed modal components to construct the Hilbert spectrogram as the network input； finally， the improved ECAResNet was used to extract the global and local features of the Hilbert spectra to achieve efficient recognition. The performance of the proposed method was tested using self-acquired USRP datasets， and the experimental results demonstrated that the accuracy of the proposed method was close to 100% in identifying six types of radar emitter individuals at high signal-to-noise ratios （SNRs）. Compared with the existing methods based on VMD， the recognition rate of the proposed method was improved by 5.41 and 7.93 percentage points at high SNR， and by 14.58 and 26.88 percentage points at low SNR （0 dB）， respectively. This suggests the superior noise immunity performance of the proposed model. Moreover， ablation experiments were designed to verify the effect of parameter optimization on the recognition performance. Compared with different recognition networks， the recognition accuracy of ECAResNet at all SNRs was improved by 1 and 6.9 percentage points， respectively， under the condition that the network parameters and amount of operation were not significantly different. Thus， the experimental results verified the efficacy of the proposed method in terms of recognition accuracy and noise immunity.