An Open-Set Recognition Method for Cross-Receiver Specific Emitter Identification Based on Transfer Learning and Support Vector Data Description

To address the challenge of low open-set recognition accuracy for specific emitter identification (SEI) under cross-receiver scenarios, a method based on transfer learning and support vector data description (SVDD) is proposed. The signals are detected through dual-window method and preamble correlation, and the instantaneous amplitude is taken as the input to the network; A residual neural network architecture with a loss function based on cross-entropy and Local Maximum Mean Discrepancy (LMMD) is designed to perform transfer learning on signal in source domain and target domain for feature extraction and closed-set identification; A hypersphere is constructed based on the features of signals transmitted by known emitters using the support vector data description to achieve accurate identification of unknown emitters. The experimental results of transmitted signals from 10 real USRP devices demonstrate that the cross-receiver open-set recognition accuracy of the proposed method has an improvement of about 40% compared to three existing open-set recognition methods.