An Open Set Specific Emitter Identification Method Using Deep Feature Embedded Discriminator

Specific emitter identification (SEI) can verify the identity of transmitter devices. It is playing an increasingly important role in maintaining internet of things (IoT) security by allowing legitimate devices to access the core network while preventing rogue devices intrusion. In recent years, more and more methods have applied deep learning to SEI. However, most of the existing works make the assumption of closed sets, which results in classifiers that can only identify a limited set of known devices. When closed set classifier is faced with rogue devices, it will misclassify them as legitimate devices, thereby endangering IoT security. In this paper, an open set SEI method based on deep features has been proposed, which can complete the identification of known devices and the detection of rogue devices by combining the appropriate transmitter deep feature distance measurement method and extreme value theory. Experiments on intercom dataset collected by universal software radio peripheral (USRP) show that our proposed method is better than other open set SEI methods. Furthermore, our method is more robust to noise. The proposed method can effectively solve the problem of open set SEI, and the detection of rogue devices is more accurate.