A self-expanding identification method for non-cooperative space radiation sources

The research on radio frequency fingerprint identification (RFFI) of radiation sources is usually regarded as a typical closed-set classification problem. However, in the non-cooperative space, the radiation sources are unknown, so the closed-set classification algorithm is inapplicable. To solve this problem, this paper proposed an open set recognition (OSR) method for the expert model self-extension of non-cooperative space radiation sources, which can self-expand to identify unknown radiation sources in non-cooperative space. Firstly, the open set method was used to identify the known/unknown radiation source samples, and several independent radiation source identification expert models were formed through the sample library construction. Secondly, a model decoupling federation strategy was proposed for the expert model self-expansion of radiation source identification, which ensured the online continual learning (CL) of the expert model on the radiation source samples in space, effectively overcoming the problems in the traditional radiation source identification model design, such as the inability to automatically learn and identify new radiation source samples and the vulnerability to catastrophic forgetting. Finally, the sample balance and interleaving techniques were used to enhance the specificity of expert models for fingerprint features of radiation sources to ensure the rapid convergence of expert models and maintain the high-generalization ability for specific radiation source identification. The experimental results show that the classification accuracy of the proposed method on the radiation source equipment is more than 97.8% in the signal-to-noise (SNR) environment of 5 dB and 100% in the signal-to-noise environment of 27 dB.