Research On Radar Fingerprint Feature Extraction And Recognition Method

Radar fingerprint identification is used to identify and track the radar individual of the same type by extracting the subtle characteristics of the transmitted signal,which are mainly caused by the differences in the internal components of the radar.Compared with traditional radar individual or modulation recognition,radar fingerprint information is more subtle and concealed.At present,machine learning-based radar fingerprint feature extraction and recognition methods are popular.However,existing studies are mostly based on simulation data,the model and parameter settings have large deviations,with low accuracy and poor stability in real scene.In addition,electronic warfare is more intelligent and automated,which puts forward higher requirements for the real-time and adaptability of radar fingerprint recognition.In view of the above challenges,this thesis first expounds the mechanism of radar fingerprint generation;then,researches on radar fingerprint feature extraction and recognition methods based on the measured radar data,and evaluates the effectiveness of the proposed methods.The main innovations and work are as follows:1.Aiming at the problems of poor feature validity and representativeness in the traditional method and the unsatisfactory recognition effect in the actual scene,a multi-domain feature extraction and recognition method based on variational mode decomposition(VMD)is proposed.In order to fully mine the fingerprint information hidden in the signal,the method first uses VMD to decompose the radar signal into different sub-signals,and then extracts multi-domain features of each component.In terms of classifier design,it is proposed a method of radar fingerprint recognition based on random forest.This thesis compares with other classifiers through theoretical demonstration and experimental analysis,and studies the problem of setting the number of decision trees in random forests.Finally,the influence of the number of sampling points in the pulse on the recognition effect is studied,and the proposed method is evaluated through comparative experiments.Experiments demonstrated that this method can realize high-accuracy radar fingerprint recognition based on partial intra-pulse steady-state signals.As the number of sampling points increases,the recognition rate continues to increase.Among them,3000 sampling points within the pulse are extracted for feature extraction,which can achieve recognition rate higher than 99%;the model has strong generalization ability,the algorithm is simple and easy to implement.2.In order to realize the automatic feature extraction and recognition of and improve the real-time and universality of the model,a radar fingerprint recognition method based on convolutional neural network was proposed.By transforming the time-domain signal into a two-dimensional time-frequency diagram and inputting the convolutional neural network for radar fingerprint features extraction and recognition.Aiming at the problems of poor time-frequency aggregation and frequency resolution of traditional methods and easy loss of fine radar fingerprint information,a time-frequency analytical method based on variational mode decomposition and Hilbert transform(VMD-HT)is introduced.In order to reduce the amount of calculation and ensure the integrity of the fine fingerprint information,researched on the preprocessing of time-frequency diagram.In view of the particularity of time-frequency diagram,the parameter settings of convolutional neural networks,the effect of time-frequency diagram size,network structure optimization and the number of convolutional layers is studied.Research based on measured data shows that the optimized network has a high accuracy and generalization ability.