Radar Emitter Recognition Based on RESR Network

Deep learning have made plenty of achievements in the field of radar emitter recognition. Radar emitter signals are typically one-dimensional. Two-dimensional convolutional neural network (CNN) methods for radar emitter signals recognition require a lot of storage, and recurrent neural network (RNN) models require a lot of training time due to their structural constraints. In order to solve the problems previously described, a method named Radar Emitter Signals Recognition (RESR) network is proposed in this paper. The most unique of our method is the RSformer blocks. RSformer blocks exploit the advantages of Transformer and CNN in capturing features and learning deep features of input signals. The approach realizes the recognition of radar emitter signals without complex data preprocessing and improves the shortcoming that RNN cannot compute in parallel. The experiment results indicate that the proposed method can achieve 90% accuracy at −14dB and 100% accuracy at −4dB. Compared with other deep learning methods, our model has a better recognition performance with higher accuracy at low SNRs.