Research On Radar Emitter Identification Using Small Training Set

With the continuous emergence of new-system radars,and electronic microwave environment is becoming more and more complex.Using deep learning methods to identify individual radar emitters can eliminate the complexity and error of artificially screened features.Its powerful feature extraction capabilities can be effective for identifying radar emitters' subtle features automatically.This article focuses on the task of identifying individual radar emitters under small sample conditions.Based on Transformer model,Deep convolutional neural networks,Transfer learning training and multi-modal feature fusion,we explored multiple deep learning methods to identify individual radar emitters.The main work and innovations of the thesis are as follows:First,for one dimensional radar signal sequence,based on the structure of Transformer model,we built an individual radar emitter recognition model.Enlightened by the idea of imitating natural language processing,embedding the radar signal to generate signal-embedding,and the radar signal is mapped to the high-dimensional hidden layer space,and using data enhancement and data expansion,to some extent,solve the problem of limited training samples.The comparison experiment proves that the model can effectively extract the subtle features of individual radar emitters,and is not interfered by conditions such as radar pulse parameters and sampling ways.Second,For the two-dimensional radar image,using transfer training method to train a deep convolutional neural network to extract features and classify specific radar emitter.In order to improve the generalization and robustness of the radar emitter individual identification model,and resist the interference of the environment and equipment.Convert the individual radar emitter signal into a radar bi-spectrum image through high-order spectrum analysis.Effectively enhance the generalization ability of the recognition model,by transfer training,in less training time for the accuracy of the measured radar emitter individual signals up to 99%.Third,in the case of better adoption to the engineering application scenarios,we built an end-to-end multi-modal fusion radar emitter individual recognition system.In order to make full use of the measured individual signal data of small-sample radar emitters,design a parallel dual-model feature extraction network to extract multi-modal fusion of individual radar emitter signal features.Using feature layer fusion and decision-making layer fusion two strategy for the sequence and image modal feature fusion,and add the heated-up softmax layer to enhance the generalization ability of the model,which reduces the parameter amount and process complexity of the model while ensuring the recognition accuracy.