Research On Intelligent Recognition Technology Of Radar Active Jamming

Electromagnetic spectrum combat has become an important component of modern warfare.Countries around the world have invested a lot of research resources to compete for the initiative in the electromagnetic space battlefield.Electromagnetic warfare is an important part of electromagnetic spectrum warfare.Radar is an important combat object in electromagnetic warfare.The problem of radar jamming and anti-jamming has always been a hot research topic.For the radar defender,the realization of accurate jamming recognition is the prerequisite for radar anti-jamming.However,traditional jamming recognition algorithms based on feature extraction have poor generalization ability and slow update of the feature database,and cannot adapt to the modern complex electromagnetic environment of the battlefield.Therefore,it is urgent to propose an intelligent jamming recognition algorithm.Aiming at the problem of radar active jamming identification,this paper introduces deep learning into the field of interference identification,and studies the jamming identification problem of small samples and low tag rates,and achieved good results.The specific research content is as follows:Mathematical modeling of six kinds of radar active jamming,including interrupted-sampling directly jamming,is carried out,and the characteristics of time domain and frequency domain as well as the jamming effect on radar are analyzed.In view of the large amount of input data,too many network parameters,and difficulty in engineering implementation caused by direct application of deep learning to jamming recognition,this paper proposes four methods to visualize one-dimensional radar jamming signals as two-dimensional images.Four signal visualization methods were simulated,and the characteristics of various jamming images after visualization were analyzed.It is proved that the four signal visualization methods can better represent the characteristics of radar jamming signals and can be used as data sets for deep learning networks.Based on the simulation scenario of the tracking radar,a simulation signal data set was created by setting reasonable jamming parameters.Based on the characteristics of the four signal visualization methods,two data preprocessing methods are proposed,and four image data sets are generated respectively as training and test data sets for deep learning.Then this paper proposes the network model based on the Xception.It trains and tests four image data sets,and compares and analyzes the results obtained with those based on convolutional neural networks(CNN).It proves that compared with the CNN model,the network proposed in this paper has achieved better recognition results on radar jamming data sets,and and the recognition results of the two networks are better than the traditional ones.Finally,based on the actual measurement data,four actual measurement data sets are generated,and the network based on Xception design is used for verification,which proves the actual effectiveness of the algorithm proposed in this paper.In view of the problem that labeled samples are difficult to obtain in actual situations,this paper combines the temporal ensemble algorithm with virtual adversarial training,and proposes TE-VAT jamming recognition algorithm.The simulation verification is carried out on the small sample data set with the number of labels of 70,105,140,and the results of the supervised learning algorithm are compared.Compared with the supervised learning algorithm,the semi-supervised learning algorithm proposed in this paper has achieved better results.At the same time,the results of the small sample data set generated by the four visualization methods are compared,and it is proved that the data set based on the short-time Fourier transform can achieve better results on the small sample data set.