Fioma: Towards Open-Set Semi-Supervised Specific Emitter Identification

Abstract

Addressing the challenge of limited labels in Specific Emitter Identification (SEI), various Semi-Supervised Learning (SSL) approaches have been explored. However, these methods often neglect pervasive unknown signals in the real electromagnetic environment, thus restricting the model’s practical adaptability. To tackle this issue, we introduce Open-Set Semi-Supervised Learning (OSSL), a novel approach that excellently enhances the model’s robustness and generalization performance in complex electromagnetic environments.We present Fioma, an OSSL-SEI model distinguished by its innovative CenterOvA Loss. This loss function combines generative (Center) and discriminative (OvA) losses to optimize feature representations. Specifically, generative loss refines intraclass distributions through labeled samples, while the Exponential Moving Average (EMA) dynamically adjusts class radii for adaptability and stability. Discriminative loss considers inliers and outliers, achieving tight intraclass clustering and clear interclass separation. Through the innovative CenterOvA Loss, Fioma exhibits notable classification and clustering capabilities, even with limited labels.