Recognition Of Radar Work Mode Based On Auto-encoder And Gradient Boosting Model

As an important part of electronic warfare,electronic reconnaissance plays an important role in percepting battlefield situation,maintaining self-defense capabilities and preventing unexpected attacks.In local space electronic countermeasures,radar work mode recognition is the basis of threat warning and accurate interference,which is of great significance for the local battlefield situation awareness and operational deployment.This paper focus on work mode recognition of airborne phased array radar.First,we simulates the pulse description words intercepted by electronic intelligence machine in different scenes and working modes.Then the amplitude of each pulse group is fitted with joint parameter modeling.Finally,the characteristics of the fitting amplitude are extracted and classified by deep neural network and gradient boosting model.The main work includes:1.According to the basic principle of radar,four typical work modes of phased array radar are modeled,and then we simulate the pulse description words intercepted by the electronic intelligence machine under different scenes and work mode,which can be used as the input of the subsequent pulse amplitude fitting and the work mode recognition.2.This paper proposes joint-parameter modeling,in this method,a single pulse is divided into different pulse groups according to the membership degree,and the amplitude of each pulse group is fitted to get the antenna scanning amplitude.The method effectively overcomes the influence of dropped pulse and false pulse.3.In this paper,a work mode recognition method based on deep sparse auto-encoder and gradient boosting decision tree is proposed.The former is used to extract the features of the fitting amplitude without supervision,and the latter is used to classify the features based on regression tree.In experimental section the influence of the sparsity parameter of the deep sparse auto-encoder and the characteristics of different hidden layers on the recognition results are analyzed.Also the influence of the maximum depth and the maximum iteration number of the decision tree in the gradient boosting decision tree on the classification accuracy is analyzed,and the classification performance of different deep model and classifier is compared.4.In view of the noise sensitivity of deep sparse auto-encoder and the low training efficiency of denoising auto-encoder,work mode recognition method based on the marginalized denoising sparse auto-encoder is studied,and XGBoost is used as the classifier to further improve the classification precision and training efficiency.