基于MAE及改进CRN的跨时间域辐射源个体识别OA

In order to solve the problems of weak model generalization ability and low signal recognition accuracy of existing emitter identification methods in cross-time domain scenarios,an emitter identification method based on a masked autoencod-er(MAE)and an improved Compression Residual Network(CRN)is proposed.The emitter identification effect is improved through a two-stage processing strategy:In the first stage,MAE with Vision Transformer(VIT)architecture is adopted.The random masking pre-training is performed on unlabeled data to infer global characteristics of the signal from local spectral in-formation,and the attention mechanism is utilized to establish effective correlations between them,thereby enhancing the model's generalization ability.In the second stage,the pre-trained VIT encoder is combined with CRN as the backbone net-work.Labeled data is used to fine-tune the model,capture more critical subtle features of the signal,learn more structured latent representations of the data,and improve the overall representation ability and recognition performance.Experimental results show that the proposed method improves the signal recognition rate of 6 types of radio emitters by 3.81%-16.18%compared with existing VIT-based methods under four different time batches.