To address the issues of low robustness and recognition performance in individual recogni-tion algorithms for unknown emitter sources within open-set scenarios,a multiple Gaussian residual prototype network is proposed.This network recognizes grayscale grophs of signals by integrating the fundamental concepts of residual networks with multiple Gaussian prototype learning.Constraints are generated to ensure that potential features of the same class cluster tightly around their respective Gaussian prototypes,thereby expanding the storage capacity for unknown class samples.Additionally,discriminant constraints are applied to increase the separation between Gaussian prototypes of differ-ent classes,enhancing the classification and discrimination capabilities of known classes.Experiments involving seven types of radiation sources demonstrate that the proposed algorithm outperforms other algorithms in terms of recognition performance under the same signal-to-noise ratio(SNR)conditions.Notably,in open-set scenarios,it achieves an AUC value that is respectively 0.207 and 0.221 higher than those of the compared algorithm,highlighting its superior recognition and classification abilities.
specific emitter identificationopen-set scenariomultiple Gaussian residual prototype networkvector graphics
维普
万方数据
