Abstract
Radio Frequency Fingerprinting (RFF) exploits the unique characteristics of device hardware and has become a key technology in IoT device authentication and network security. Identification of unknown devices is a key challenge for radio frequency fingerprinting (RFF) in open-set scenarios, the similarity of device hardware characteristics further exacerbates the difficulty of the task. This paper proposes an open-set RFF recognition framework called SV-NPR (Siamese VGG16 with Negative Prototype Rejection). The framework combines the advantages of the VGG16 network in local feature extraction with the contrastive learning mechanism of the siamese network, and can efficiently capture the distribution of local detail features in RF signals. In addition, the introduction of a dynamic rejection mechanism based on negative prototypes improves the robustness and generalization ability of the model for unknown categories. Experimental results show that SV-NPR significantly outperforms the state-of-the-art on the Oracle dataset and exhibits leading recognition capabilities in open-set scenarios.