Radar Emitter Signal Recognition Algorithm Based On Deep Learning

In electronic countermeasures,radar emitter signal recognition plays a crucial role in the pattern of war.In recent years,radar emitter signal recognition technology has made great progress.However,the traditional identification methods have been unable to meet the needs of automation,informatization and intelligence in electronic warfare due to the large amount of calculation and the increasingly complex electromagnetic environment.Therefore,this dissertation proposes a radar emitter signal recognition algorithm based on deep learning,which uses the excellent image feature extraction ability of deep learning structure to extract and classify the signal time-frequency image.The specific work is as follows:Firstly,the basic concepts and mathematical models of radar emitter signal are introduced,and each signal is simulated.The time domain and frequency domain waveforms of the signals are compared and analyzed.Then it introduces two kinds of time-frequency analysis of short-time Fourier transform and Wigner-Ville time-frequency distribution,and gives the time-frequency transform pictures of six kinds of signals after short-time Fourier transform.The digital image processing technology is introduced to preprocess the six kinds of signals.Secondly,in order to improve the recognition performance of radar emitter signal in the case of low SNR,the radar emitter signal recognition algorithm based on stack denoising auto-encoder(sDAE)and convolution neural network(CNN)is studied.Simulation results show that the two deep learning structures can achieve better recognition results than other feature extraction algorithms under low SNR.Compared with the two algorithms,the sDAE can mine deeper features of the input data,and has better recognition performance,and has good dimensionality reduction performance;CNN can greatly reduce the network computation through local connection,weight sharing and pooling layer operation,using this model has lower time complexity.Thirdly,in the case of high SNR,a DAE + CNN recognition model is proposed to reduce the time complexity on the basis of guaranteeing the recognition performance.Firstly,the input data is nonlinear dimensionally reduced by DAE,and then the dimensionally reduced data is classified and identified by CNN.The simulation results show that the algorithm model achieves ideal recognition effect under the condition of high SNR,and the time complexity decreases obviously compared with sDAE and CNN,which can be applied to high real-time scenarios.