Research On Individual Recognition Of Radar Emitter Based On Convolutional Neural Network

With the introduction of new technologies related to electronic warfare and the increasingly complex electromagnetic environment,electronic warfare processing has the possibility of further development,especially in the military field that requires accurate identification and analysis of radar signals.At present,radar emitter signals can be recognized according to the difference of pulse descriptors and the different types of intentional modulation in pulse.However,as the electromagnetic environment becomes more and more complex,it is difficult to complete the correlation identification and analysis of radar signals only by using traditional methods.At the same time,different radar transmitters have different internal devices,these internal differences are called unintentional modulation information,so signal recognition can be realized through this part of information.Although the intra-pulse unintentional modulation has unique features among different individuals in complex electromagnetic environment,but the extraction of these features requires complex formula reasoning and the feature extraction process is complex and difficult.In view of the above problems and challenges in the classification and identification of individual radar emitter signals,this article is based on the research foundation of the predecessors,combined with the relevant research of convolutional neural networks in some fields,and used it to design the individual radar emitter Identified network,and mainly conducted the following research:(1)Since the input signal is a one-dimensional signal,a one-dimensional network structure is designed using a convolutional neural network to complete the signal classification and recognition.Taking into account the difficulty of signal feature extraction,multiple convolution kernels of different sizes are used for convolution to complete the automatic extraction and analysis of relevant features to achieve the identification of different signals.And considering problems that there may be small feature differences between different signals,there is a large of feature differences between the same signal,loss function is optimized to further improve the classification effect.A large number of experimental results show that the convolution neural network based on one-dimensional convolution structure can achieve better individual feature extraction and discrimination of radar emitter signal,and have good effect in radar individual recognition.(2)The network structure based on one-dimensional convolution design confirms the important role of pulse internal characteristics in radar emitter identification.In order to be able to more accurately identify different types of radar individual signals,given the significant effect of convolution neural network in image classification,according to the fingerprint characteristics of intra-pulse modulation has no intention of,can be used as a unique characteristic of radar emitter individual,consider using a two-dimensional convolution neural network,the realization of radar emitter signal recognition of individual pulse envelope image characteristics.Firstly,a pulse envelope extraction algorithm is designed to extract the pulse envelope of radar emitter individual signal.Then a network based on attention and dense block is proposed for pulse envelope feature recognition.Attention can better learn the key information between different fingerprint features and complete the update of the weight of the feature map in the channel dimension.Dense block can fuse all the feature information in the network and effectively complete the reuse of feature information.Finally,the experimental results show that the two structures can effectively extract the feature information of in pulse unintentional modulation in complex electromagnetic environment and have good performance in noise immunity,which can effectively complete the individual recognition of radar emitter signal.