Data Enhancement Technology For Individual Recognition Of Small Sample ADS-B

The electromagnetic spectrum space has become one of the country’s new security areas,a strong support for the formation of joint combat capabilities based on network information systems,and a key component of global combat capabilities."Control of the electromagnetic spectrum" will occupy the strategic commanding heights in the future battlefield,and the identification of electromagnetic individual will also become the key to victory in information warfare.At present,there are two key problems in the research of electromagnetic individual recognition: The first is the increasingly complex electromagnetic environment of the battlefield in information warfare,and traditional technical methods to deal with it have been difficult to quickly and effectively identify electromagnetic individual;The second is that the deep learning network model requires a large number of data machines for training.However,in most practical scenarios,due to the fast acquisition of electromagnetic signal equipment and the abstract electromagnetic signal waveform itself,it is difficult to establish a large number of high-quality labeled data sets,which causes greater difficulties in electromagnetic individual recognition methods.Aiming at these two major issues,this paper studies the small sample electromagnetic individual recognition technology based on deep neural networks.The main research content is the following two aspects:First,this paper studies the electromagnetic individual feature extraction and implicit knowledge discovery technology based on deep complex network.This paper studies the related information extraction mechanism of electromagnetic individual I\Q waveform domain signal based on complex number operation,reduces the degree of freedom of weight,and makes the recognition model better and faster to fit the implicit features of electromagnetic individual.Secondly,this paper studies the time series data enhancement method based on the data level.According to the characteristics of fading,multipath delay and Doppler effect in the propagation of electromagnetic signals,this paper designs a time sequence transformation enhancement method that meets the characteristics of electromagnetic individual signals:noise disturbance,amplitude transformation,time delay transformation,frequency offset transformation and phase offset transformation.This paper verifies the effectiveness of the time-series transformation method in this paper under the parameters of data enhancement multiples,different signal-to-noise ratios and data enhancement positions through the measured small sample ADS-B data set.Finally,this paper studies the virtual confrontation training enhancement method based on the model level.This paper studies the generation mechanism of virtual countermeasure training,and realizes the application of data enhancement in electromagnetic individual recognition based on virtual countermeasure training.This paper proposes a method of combining virtual confrontation training and time series transformation,that is,applying regularization methods at the model level to obtain better model parameters,and at the data level,expanding the individual features required for network recognition.