Research On Radar Emitter Individual Recognition Technology Based On Deep Learning

With the continuous development of information technology and the fact that moden battlefield electromagnetic environment is becoming more and more complex,radar,as an important equipment of battlefield perception,receives a variety of complex electronic warfare field emitter signals.How to utilize such a huge amount of high-dimensional,redundant,complex,strongly interfered and unevenly distributed radar radiation signajcs to accurately identify the target is the key problem of electronic countermeasures.However,traditional artificial feature extraction or machine learning methods are difficult to extract effective features with high discmminability from such complex radar radiation source signals.In addition,in the actual battlefield environment,there are still some key problems to be solved urgently,such as model deployment and identification of unknown radiators.Based on the strong feature extraction ability of deep learning,this study proposes a deep learning based radar emitter individual recognition model according to the characteristics and difficulties of radar data and the problems in complex battlefield environments.The main work and contributions of this study are as follows(1)By analyzing the characteristics and difficulties of individual data of radar radiation sources and comparing with traditional machine learning methods,this study explains the concept of "capacity1" of model and problem,and proposes a deep learning based individual recognition model of radar radiation sources under technical optimization.At the same time,aiming at the problem of task adaptation and platform adaptation of the actual battlefield model development,an individual radar emitter recognition model based on a neural network structure search algorithm,called ProxylessNas,is proposed.(2)According to the characteristics of high redundancy,strong interference and limited effective signal range of radar radiation source individual signal,two kinds of attention mechanism of radar emitter "fingerprint" signal location,feature extraction and individual recognition model are proposed,making the model focus on the discriminative range of data and extracted features,mining "fingerprints" of radar emitter signal and learned features.Furthermore,the accuracy of radar radiation source recognition is slightly improved.(3)In order to solve the problem of identification of unknown radar radiation sources which are caused by various kinds of similar radar radiation sources,endless emergence of new radar models and limited database capacity in actual battlefield environment,the model of radar radiation source individual recognition and unknown radiation source identification is proposed based on deep metric learning,which can recognize the known radiation source and identify the unknown radiation source at the same time and make the identification accuracy of radar unknown radiation source further improved.