Research On Accurate Recognition Method Of Radar Radiator Signal Based On Deep Learning

Radar radiator signal recognition is a key research field of radar countermeasures.Obtain enemy radar technology and electronic intelligence by cracking the information carried by the signal,ensuring the military status on the battlefield and effectively directing operations,Accurate recognition of radar radiation source signals,accurate judgment of radar working mode,and correct identification of individual radars are the core points of radar countermeasures and reconnaissance systems.The emergence of various new systems of radar has made the radar environment more complicated,and for anti-jamming and antireconnaissance,the modulation of the radar source signal has become more complex and the parameters changeable.Therefore,under the condition of low signal-to-noise ratio,inpulse recognition of radar radiation source signals,work pattern recognition and individual recognition have become an urgent problem to be solved.This article focuses on the application of deep learning in the signal processing of radar emitters.The specific work content includes the following aspects:1.Aiming at the shortcomings of traditional algorithms for poor signal recognition and lack of real-time and high efficiency,two deep learning-based radar signal recognition methods are studied.Using the mutual ambiguity function based on linear canonical domain and linear canonical transformation,method one extracts the cross-sectional maximum value of the mutual ambiguity function to obtain two-dimensional feature information.In order to solve the problem that manual feature extraction consumes a lot of time and cost,a recognition method based on deep sparse capsule network is proposed.The simulation results show that the method in this paper effectively reduces the time cost and has good recognition performance.The second method uses Choi-Williams distribution to extract the time-frequency diagram of the radar signal.Reduce the dimensionality of the feature image.The input stack is mixed from the encoder to extract the features,and the semi-supervised linear discriminant method is used to reduce the dimensionality of the features to further extract the classification features.The degree of discrimination of the same category is reduced as much as possible,and the degree of discrimination of different categories is increased as much as possible,so that the features learned from the innermost hidden layer of the encoder are more conducive to classification.The simulation results show that this method has good noise immunity and a high recognition rate under low signal-to-noise ratio.2.Aiming at the difficulty of working mode recognition in the pulse stream where the number of working modes is unknown and the duration of each working mode changes irregularly,a radar working mode recognition method based on deep learning is studied,and the PDW characteristics and amplitudes of radars are combined by canonical correlation analysis.After fitting the frequency domain features,the deep confidence network is introduced into the field of radar work pattern recognition for recognition.The simulation results show that this method can get a better recognition rate.3.Aiming at the problem that the individual radar fingerprint features are greatly affected by noise,and the stability and adaptability are relatively poor,a radar individual recognition method based on deep learning and feature fusion is studied.Using pulse coupled neural network,the radar time-frequency domain characteristics and modulation domain characteristics are fused.And use the deep multiple wide residual network as the classifier to realize the recognition function.First,the representative slice of the ambiguity function and the rectangular integral bispectrum are extracted as the primary features.Pixel-level fusion of features is performed through PCNN.Finally,the features are input into the deep multiple wide residual network to realize accurate individual recognition.The simulation results show that the method overcomes the instability of fingerprint features under low signal-tonoise ratio and has good recognition performance.