Radar Signal Recognition Based On Deep Belief Network

In today's world,the development of science and technology is changing with each passing day,and every area is constantly improving.Radar technology as an important part of the military struggle is also constantly developing.A new system radars have been universally deployed to the armies of various countries.The electromagnetic environment of the battlefield in the future has become very complicated,which has brought a lot of challenges to electronic reconnaissance.Traditional reconnaissance methods have obviously failed to identify the intrapulse modulation of signals.In order to solve these problem,this paper introduce the deep learning algorithm into radar emitter signal recognition.The whole process is divided into three parts: signal preprocessing,signal feature extraction,and identification and classification.1.Analyze the blind source separation algorithm for mixed signals.In Chapter 3,we will study the separation of blind signals to a certain extent.In the simulation process,five different modulation signals are set.The five signals overlap in the time domain and frequency domain,according to the transient linear mixed model.The mixed signal was separated using the method named Fast ICA.The results showed that the five mixed signals were successfully separated.2.Propose a method of radar emitter identification based on deep belief network.After preprocessing the signal,then convert time domain signals to time-frequency domain by using time-frequency transform method.In order to simplify the feature information and improve the accuracy of the classification,the quadratic feature is extracted from the timefrequency matrix.So we extract the singular value spectrum from time-frequency matrix as the basis for classification.Last but not least we use the deep belief network as a classifier based on features.By adjusting the parameters of deep belief network,a better classification result is achieved.When SNR is 0d B,recognition accuracy reaches 95%.3.A multi-feature fusion recognition method is proposed.In order to solve the problem of low recognition rate when signal contains a lot of sales,this paper presents several methods for multi-features fusion.Include singular value spectrum joint classification,timefrequency matrix combined to classify,and time-frequency matrix fusion to classify.Verified by simulation,each method can be effectively identified.When SNR is-5d B,the recognition accuracy has been improved.Prove that multi-feature fusion method is effective.