Research On Feature Extraction And Recognition Of Low-frequency Radiation

In the field of communication countermeasure and electronic reconnaissance,the identification of low-frequency radiation source is an important research topic.Even if the type of radiation source and the type of transmitted signal are known,it is difficult to achieve full replication.This is because there are inevitable differences between the various components that make up the radiation source,and these differences are different from each other.In this paper,the method of feature extraction and classification recognition of low-frequency radiation sources is studied.The main research contents are as follows:First of all,to find an effective signal processing method,only on the basis of finding a reliable signal processing method can we extract stable and distinguishing feature.This paper mainly studies two time-frequency analysis methods,local mean decomposition(LMD)and variable mode decomposition(VMD),and compares the decomposition effect.After the decomposition step,the marginal spectrum is extracted,combined with chirp Z transform(CZT)without increasing the number of sampling points and the amount of computation.This transform greatly improved frequency resolution and realized spectrum refinement.Secondly,the feature extraction method based on time-frequency analysis is studied.On the basis of the spectrum refined marginal spectrum,Shannon entropy and index entropy as well as the correlation dimension features of the signal are calculated,and the simulation signal and the measured signal are analyzed respectively.The analysis results show that the combination of these three features can achieve better separability.Then,we study the extraction of bispectral features,introduce the theory of spectrum,and propose a method to cast the signal from the high-order spectrum domain to the eigenvalue domain by using the 2D image Fourier transform,and on this basis,extract spectral features such as spectral brightness,spectral flatness and spectral roll off.In addition to the basic graph Fourier transform,the weight in the transform is also adjusted.For bispectrum concentrated in the low-order region,the new definition of the weight can highlight the difference in the low-order region.The analysis results of simulation and measured signals show that the method improves the effectiveness of spectral characteristics.In the last stage of classification and recognition,support vector machine(SVM)with kernel of radial basis function(RBF)and Bayes classifier are used to classify and recognize the features extracted above under different SNR.The results show that the improved combined feature extraction method can effectively distinguish the radiation sources.