Application Of Machine Learning In Individual Recognition Of Radiation Source

In the increasingly complex electromagnetic environment,the identification of individual radiation source is an important means to obtain information in electronic warfare.The"hardware"fingerprint identification of the radiation source individual refers to the ability to associate the unique electromagnetic characteristics of the radiation source with the radiation source individual,also known as the fingerprint identification of the radiation source signal.However,at present,the"fingerprint"recognition rate of hardware based on traditional classification and recognition of radiation sources is low and the classification performance is poor.Therefore,this paper proposes a radiation source hardware"fingerprint"recognition method based on machine learning and deep learning.This topic from the extract characterization of emitter individual hardware"fingerprint"the characteristic,the analysis of the antenna far area of field intensity distribution and the directivity of the spatial distribution,and other"hardware"fingerprint characteristics,in-depth analysis of the principle of a variety of machine learning model,realize the use of machine learning algorithms identify individual"hardware1" of the fingerprint characteristics of radiation source,and achieved a high recognition rate,has great practical significance.The main contents of this paper include:1.Analyze the hardware characteristics and directional parameters of radiation source individuals,and complete the hardware"fingerprint"feature extraction of radiation source individuals through data mining method,including the remote field intensity distribution of radiation source and the spatial distribution of various directional parameters.2.According to the SVM model related theories,parameter design is carried out to construct an appropriate SVM classification model,and several metrics of the SVM model are given.The optimal parameter combination is obtained by comparing the effects of the model parameters on the recognition rate.3.For the case that the method of single machine learning model can easily lead to overfitting,the random forest model with integrated machine learning method is proposed.Based on the statistics of the importance of each feature,the random forest classification model is constructed.The optimal parameter combination is obtained by comparing the effects of the model parameters on the recognition rate.4.In order to further improve the recognition rate and pave the way for more complex electromagnetic problems,a neural network classification model is proposed for the"hardware"fingerprint recognition of radiation sources.