Few-Shot Specific Emitter Identification Based on Fuzzy Oversampling Data Augmentation

Specific emitter identification is a key technology for spectrum management and electromagnetic environment awareness. Data privacy protection makes it challenging to obtain labeled data of specific emitter, and the few-shot problem causes the overfitting and decreased recognition performance in deep learning methods which depend on large datasets. This paper proposes a method based on Fuzzy Oversampling Data Augmentation (FODA), which generates synthetic samples in random directions and assigns multi-class fuzzy labels to achieve diversified sample augmentation. And the intra-class compactness of samples is enhanced by reducing the variations between semantic features and corresponding class centroids in the high-dimensional feature space, leading to optimized classification performance. Few-shot experimental scenarios are constructed with a 10-class Wi-Fi public dataset. The experimental results indicate that the identification accuracy of FODA has achieved optimal average accuracy compared to the comparative models across different shot data, and been improved on average by 24.61% compared to the baseline in the ablation experiments.