Research Of Radar Working Mode Recognition With Few Samples

As an important part of the radar countermeasure system,radar signal recognition plays an unparalleled role in electronic reconnaissance.The recognition of radar signals includes radar signal sorting and radar working mode recognition.Radar signal sorting is to separate randomly overlapping pulse streams into separate pulse trains for each radar;radar working mode recognition is to identify the working mode of each radar based on the results of sorted pulses.Only by recognizing the working mode of the opponent's radar can it predict its behavioral intentions and carry out correct interference.The thesis focuses on the research of radar working mode recognition task.The more information we obtain from the signals of radar radiation sources,the more support we can provide for our actions.However,in actual situations,the number of pulse samples received from the radar signal of the adversarial may be very small,which makes us face many difficulties in identifying the working mode of the radar radiation source.Traditional radar work mode recognition methods are based on the premise of having a large number of training samples,such as support vector machine algorithms based on machine learning and convolutional neural networks based on deep learning.In the above methods,when the number of training samples is small,the accuracy rate will drop and the generalization performance will deteriorate,and it is difficult to complete the radar working mode recognition task in the few samples scenario.How to complete the recognition of the working mode of the radar radiation source and improve the recognition accuracy with a small sample size is of great significance.The main work and contributions of the thesis can be summarized as follows:1.Working mode recognition of few radar samples based on multi-class support vector machine algorithm and convolutional neural network are studied,and simulation experiments are compared.The simulation results demonstrate that these two algorithms have poor recognition performance in small sample scenarios.2.Propose two small sample radar working pattern recognition methods based on prototype neural network and graph convolutional neural network,and compare it with traditional recognition methods.The comparison results demonstrate that the recognition performance of these two algorithms is better than traditional multi-class SVM and convolutional neural network.3.The thesis proposes an encoding method that takes the value of radar working mode parameters and its changing rules as prior knowledge.The parameters include pulse repetition frequency,pulse carrier frequency and pulse width.4.A method is proposed to integrate the prior knowledge into the prototype neural network and graph convolutional neural network to assist small-sample radar work mode recognition.Simulation results demonstrate that,the recognition accuracy of the two networks increased by 3.7% and 8.7% respectively by using prior knowledge.It also proposes how to use the above two networks to complete the task of small-sample radar working mode recognition when the prior knowledge is inaccurate and part is missing.