Research On Specific Radar Emitter Identification Technology Based On Deep Learning

Specific radar emitter identification(SREI)is a process of inferring the individual information of the emitter by analyzing the signals intercepted by the reconnaissance receiver,so that the information of the emitter can be obtained.Radar emitter identification is usually based on a set of characteristic parameters of radar signal.In recent years,with the rapid development of artificial intelligence technology,deep learning has been widely studied by scholars.Through deep learning network,features which cannot be obtained by traditional methods can be extracted,and recognition performance can be improved accordingly.Based on the existing SREI technology using deep learning,two improved algorithms are proposed to further improve the recognition performance.The main work of this thesis is as follows:1.The current research status of SREI based on deep learning and few-shot learning is described firstly.Then the radar transmitting signal models are introduced.It is also introduced that the basic structure and training mechanism of deep learning network,and the structure of Convolutional Neural Network(CNN)and Generative Adversarial Networks(GAN)in detail.2.An improved algorithm of SREI based on feature enhancement is proposed.Firstly,the radar pulse signal is transformed by short-time Fourier transform(STFT)to obtain the time-frequency image,then the time-frequency image is pre-processed and input to sparse auto encoder(SAE)for feature extraction,and the extracted features of SAE are enhanced with a histogram equalization method.Finally,the verification experiments based on stimulation signals and real acquisition signals are carried out.The experimental results show that the identification performance of the proposed algorithm is greatly improved after the feature enhancement,and the recognition performance on the basis of the real data has averagely 9 d B better than that of the algorithm without feature enhancement.3.Based on the aforementioned algorithm,a further improved algorithm with few-shot learning is proposed.Two schemes are proposed.One utilizes the GAN module after the SAE module,and the other applies the GAN module after the feature enhancement module.The simulation results show that the recognition performance of the two schemes is greatly improved.The recognition performance of scheme one is averagely improved by 1d B,while that of scheme two is improved by 15 d B on average.In addition,the number of samples required by scheme two only needs 25% samples of the algorithm without improvement on the premise of reaching the same recognition rate.