The Classification And Recognition Of Radar Radiation Source Signals Based On Deep Learning

The recognition of radar radiation source signals is a key link in electronic warfare,with nearly half a century of development,the recognition technolog of radiation source signals has made great progress.However,due to the increasing complexity of the electromagnetic environment and the specificity of the passive confrontation identified by radiation source signals,the traditional signal analysis method based on the conventional parameters has been far from unable to meet the requirements of identification,the existing pulse feature analysis method generally extract the characteristics through man-made means,and the calculation is relatively large,it is difficult to meet the electronic reconnaissance systems and equipment needs for automation,informatization and intellectualization.Therefore,this paper studies a classification and recognition method based on deep learning.The algorithm can independently learn the characteristics from the data,and combine the low-level features to form more abstract high-level representation or the attribute categories or features,and has achieved good results in the intelligent classification of images.In this paper,we discuss the popular method of deep learning neural network in recent years to solve the classificition of radar radiation source signals.Through the time-frequency conversion of several classic radar signals,the obtained time-frequency distribution maps are processed by filtering,graying and so on,and then are input into the deep neural network,then implementing the recognition algorithm of radar radiation source signals based on deep learning.The main work of this paper is as follows:1.The paper simulates the signals of six common intentional modulation signals,conventional pulse signals,linear frequency modulation signals,phase code signals(biphase coded signals and quadrature phase coded signals)and frequency code signals(dual frequency coded signals and four frequency coded signals).The Margenau-Hill time-frequency distribution and Wigner-Ville time-frequency distribution are made for signals,and the resulting time-frequency distributions are further processed to obtain training and testing data of deep network.2.Utilizing automatic encoder network to classify and recognize signals,analyzing the network parameters,learning rate,block size,training times,to obtain the best signals recognition rate under the network.At the same time,we compared the classification results of different sample sizes and prove the relationship between the performance of the automatic encoder and the number of samples.3.Establishing deep belife network applicable for classifying and recognizing radar radiation source signals,similar to previous work,the network parameters are analyzed and the classification results are compared under different parameters conditions,proving the relationship between the performance of the deep belief network and the number of samples.4.Introducing the convolution neural network,complying the classification and recognition of radar signals based on the convolution network.The convolution neural network uses local connections,weight sharing,pooling strategies,greatly reducing training parameters.Experiments test the relationship between the performance of the convolution neural network and the number of samples.At the same time the above three networks all have preferable anti-noise performances.