Data Augmentation and Individual Identification Method for Emitters Using Contour Stella Image Mapping

In wireless communication networks (WCN), numerous devices are interconnected with one another, which poses challenges to the security of WCN. Specific emitter identification (SEI) technology is considered a crucial technique for ensuring the security of WCN, which plays a paramount role in resisting network attacks and enhancing system security. However, the current SEI methods are heavily reliant on a considerable amount of labeled data, which results in difficulties in real world wireless communication situations where obtaining such data is arduous and impracticable to fulfill the application demands. To address this problem, this study proposed a few-shot SEI method based on data augmentation of contour stella image (CSI-DA). First, it utilizes the contour stella image (CSI) for data feature visualization, and then employs a generative adversarial network with patch discriminator (PGAN) for data augmentation. Subsequently, the generated images were filtered using structural similarity calculations to eliminate samples that exhibited significant differences in feature distribution compared to the original image, thereby improving the quality of the generated data. Finally, classic ResNet was employed to extract features and perform classification on both the original and generated images. The superiority of the proposed method was verified by experiments conducted on real-world datasets, achieving an 85% classification accuracy with only five samples per emitter, which was higher than existing few-shot SEI methods.