Method for imbalanced specific emitter identification based on temporal attention

To address the problems of prominent long-tailed data distribution and high-dimensional sampling features in specific emitter identification (SEI), an improved SEI algorithm based on temporal attention was proposed. By constructing a multi-scale feature fusion module to integrate feature information under different temporal granularities, and designing an identity-aware temporal attention module that generates and embeds identity-adaptive attention weights, the model was encouraged to focus on individuals with a small number of tail samples and identity-related temporal features. Experiments on the self-collected AIS dataset and public ADS-B dataset demonstrate that the proposed algorithm achieves recognition accuracies of 94.89% and 93.23% respectively, and the accuracy decreases by less than 2% under a low signal-to-noise ratio (SNR) of 3 dB.