Radar Emitter Recognition Technology Based On Deep Learning

As an important research in the field of Radar Countermeasures,radar emitter recognition technology can acquire enemy radar information and tactics electronic intelligence in order to provide battlefield situation information and decision-making action for operational commanders.With the premise of the signal sorting,the signal envelope is used as the individual characteristic to disscuss the application of deep learning methods on radar radiation source recognition.The main content of the dissertation is summarized as follows.The first part elaborates the model,strategy and algorithm in the statistical learning.In the model,this part summarizes the features and applications of the common nonlinear activation function of neural networks,and the model of the muti-classification tasks is given.Then the gradient disappearance and explosion phenomenon is analysed and a method is pointed out to relieve the problem.In the strategy,the loss functions based on mean square error and cross entropy is compared with each other.In the algorithm,the feasibility of gradient descent algorithm in deeplearning areas is discussed.Then the Momentum,RMSprop and Adam algorithm which are all improved from gradient descent algorithm are researched.The second part focuses on the models of convolution neural network(CNN),recurrent neural network(RNN)and long short-term memory(LSTM)neural network.At the same time,the application of these networks on radar emitter recognition is discussed in this part.As is known to all,the adjustment of deep learning hyperparameters affects the performance of the model,so the problems encountered in the process of network hyperparameter adjustment are summarizes and the the corresponding solutions are given.In addition,a method for reading and computing the large amount of data is proposed.At last,measured data experiment is used to verify the advantages of the deep learning method.The third part is contributed to reinforcement learning.Firstly,we introduce the basic model of reinforcement learning,and analyse the problems of strategy evaluation,strategy improvement and optimal strategy solution.Secondly,the specific description of the actual application environment in Markov decision process(MDP)is expressed.Lastly,combining with DQN,DDQN and Dueling Network,we provide a method to apply the decision-making model to the radar emitter recognition task,and then measured data experiment is used to analyse their performance.From the view of the XXX radar reconnaissance equipment,a method on specific emitter identification of radar radiation source is discussed and a parallel implementation method of deep neural network based on NXP QorIQ T4240 AltiVec engine is proposed.The real time and effectiveness of the method is verified by the measured data experiment.