Research On Individual Identification Of Radiation Source Based On Deep Learning

Electronic countermeasures(ECM)has been playing a key role in the process of fighting for the control of electromagnetic power.Electronic countermeasures cover electronic reconnaissance,electronic interference,electronic defense and other aspects,among which electronic reconnaissance is mainly to complete the identification and analysis of enemy radiation sources.Therefore,the research on individual identification of radiation source has always been a very important topic.In the traditional study of individual identification of radiation sources,the authors use unintentional modulation parameters to perform complex mathematical reasoning and extract individual characteristics of radiation sources.However,due to the complex and difficult process of extracting these features,the effect of individual identification of radiation source is not good.Therefore,in order to solve the limitations of using traditional method to extract the source signals characteristics,this thesis studied the use of deep learning techniques to complete the emitter individual identification,given the depth of learning technology requires a large amount of data in network training,but most of the emitter individual identification task scenarios are cooperation,difficult to get a lot of data samples,Therefore,this thesis also studies the enhancement of data samples by using generative adversarial network,specifically in the following two aspects:(1)One-dimensional convolutional neural network is often applied to feature learning of one-dimensional data and has achieved good results.However,considering that the radiation source signal is still a kind of time series data,this thesis tries to add short and short time memory network on the basis of one-dimensional convolutional neural network to construct a new radiation source individual recognition network.In addition,attention mechanism is introduced to further improve the recognition accuracy of the network.A series of experiments are carried out to determine the structure of the network and verify its effectiveness.Experiments show that the proposed model can improve the accuracy of individual radiation source identification to more than 95%.After determining the model structure of radiation source individual recognition,this thesis proposes a sample enhancement model based on generative adversation network to solve the problem that it is difficult to obtain sufficient sample data in practical application scenarios.According to the actual situation of individual task of emitter,this thesis improves the basic generative adversace network,and proposes a hierarchical generative adversace network,which can realize fine processing of emitter signal.The final experimental results show that when the data in the dataset is enhanced by sample,the classification accuracy of the model can be improved by 8%.