Research On Radar Signal Recognition Based On Neural Network

Radar is an indispensable part in modern battlefield.Radar emitter detects targets by transmitting high-power modulated signals,so it is particularly important to extract in pulse features of radar signals for recognition.Radar signal modulation is divided into two parts:intentional modulation and unintentional modulation.At present,classification and recognition methods based on neural network can not only automatically extract signal features,but also obtain good recognition results.According to the modulation of the signal and the characteristics of the signal itself,this paper designs a radar signal recognition system based on neural network.1.In the aspect of radar intentional modulation feature recognition,aiming at the problems of complex preprocessing of raw data and insufficient attention to noise,this paper proposes a radar intentional modulation feature recognition method combining denoising convolutional neural network and perception network.Firstly,this method models and simulates several modulation types of signals commonly used in LPI radar,and uses time-frequency transform to transform one-dimensional time-domain signal into two-dimensional image signal.Then,the convolution neural network based on residual network is used to denoise the time-frequency image,which can preserve the signal characteristics and reduce the background noise.Finally,the feature extraction of the denoised image is carried out by using the Inception-V4 network.At the same time,this network solves the problem that the general convolution neural network has too many parameters in feature extraction and the model optimization is difficult to some extent.The simulation results show that this method not only realizes the classification of different modulation modes,but also makes the network have strong robustness.2.In the aspect of radar unintentional modulation feature recognition,because radar emitter signal has strong correlation in time sequence,convolution neural network is difficult to extract the temporal logic correlation feature of data.This paper proposes a method of radar emitter individual recognition based on LSTM network and Transformer network.Firstly,the unintentional modulation model of radar emitter is established,and three signal generators are used to collect the actual data and establish the database.Then,the phase feature and envelope feature are extracted from the collected signals.Finally,LSTM network with Attention mechanism and Transformer network based on self attention mechanism are used to further extract and classify the signals.The experimental results show that the sensitivity of the two algorithms to envelope feature and phase feature is different,and the data length of a single pulse sample signal has an impact on the recognition effect of the two algorithms,but on the whole,the Transformer network model is better than the LSTM series network model in recognition effect and speed.