Research On Intelligent Processing Method Of Anomaly Electronic Reconnaissance Data With Incomplete Information

The processing of anomaly electronic reconnaissance data with incomplete information is one of the key problems in the field of electronic warfare.In the actual reception of electronic reconnaissance data,there are random missing,discontinuous,and mutation problems,where random missing and discontinuous produce incomplete electronic reconnaissance data;mutation produces anomaly electronic reconnaissance data: all of them have negative effects on target detection and target tracking.This thesis establishes 2 models of incomplete electronic reconnaissance data and anomaly electronic reconnaissance data respectively,including how to fill in the missing data of incomplete electronic reconnaissance data and propose a method of anomaly electronic reconnaissance data detection and classification.The main research results obtained so far are as follows:In view of the random missing and discontinuous situation of electronic reconnaissance data,this thesis proposes an intelligent imputation method for incomplete electronic reconnaissance data based on Bidirection Gated Recurrent Unit(Bi GRU)combined with Generative Adversarial Networks(GAN).First,the normal electronic reconnaissance data sequence and the missing data sequence are used as the input of the generator in the generative adversarial network at the same time,and the completed data is obtained after being processed by the generator,where the training of the generator and discriminator will alternate until the generator can produce data that conforms to the distribution of the original dataset.When the missing length gradually increases,simulation results show that the MeanSquare Error(MSE)value between the complete data and the original data obtained by the Bi GRU-GAN method is lower than that of the traditional method.Obviously,the modeling effectiveness is better,and the data completion speed is faster.In view of the fact that most of the current anomaly detection methods cannot extract the effective features of anomaly electronic reconnaissance data,and do not fully consider its temporal correlation.This thesis proposes an anomaly detection method based on variational autoencoder(VAE)combined with bidirectional gated recurrent unit network(Bi GRU).First,this method uses sliding window to input anomaly electronic reconnaissance data into the encoder to capture effective features,and a Bi GRU network is built to predict the sequence of the next window,and then input the anomaly electronic reconnaissance data to the decoder to reconstruct the anomaly electronic reconnaissance data.Last,anomaly detection is performed by using the calculated reconstruction probabilities compared to a given threshold.The simulation results show that this anomaly detection method can detect four common types of anomalous electronic reconnaissance data of automatic dependent surveillance-broadcast(ADS-B): random deviation injection,fixed deviation injection,Denial of Service(DOS)injection,and track replacement injection.Furthermore,this thesis proposes a classification method of anomoly electronic reconnaissance data based on Bi GRU network,which can classify four common anomoly electronic reconnaissance data.The simulation results show that,compared with the existing methods,the recognition accuracy of the proposed method is improved by 5%.