Individual Recognition Method of Radiation Source Based on Deep Subdomain Adaptation Network

Abstract

Aiming at the problem of the recognition accuracy degradation caused by the channel noise inconsistency between the signal of the radiation source to be identified and the trained radiation source dataset, this paper proposes an individual identification method of radiation sources based on subdomain adaptation. First, the radiation source signal is spliced into IQ images to generate a feature data set. Then the improved Resnet-50 network which has been pre-trained on the image set is utilised to extract the common characteristics of the source and target domains. Finally, the local maximum mean difference adaptive layer is added. By calculating the pseudo-labels in the target domain to match the conditional distribution distance, the difference in the characteristic distribution of sub-type radiation sources under different signal-to-noise ratio (SNR) can be reduced, and the model recognition accuracy can be improved. The experimental findings reveal that the proposed method greatly increases the accuracy of the individual radiation source identification technique based on the deep neural network. Under the condition that the SNR of the signal to be detected increases or falls by 4 dB compared to the training dataset, the identification accuracy of the approach suggested in this study improves by 11.4 percent and 12.7 percent respectively compared with the model Resnet-50.