Research On Radar Emitter Recognition Method Based On Convolution Neural Network

Radar emitter recognition is an important part of radar reconnaissance system.The timely and effective radar emitter recognition can obtain the information of the enemy radar's position,model and threat level,and get the first opportunity in modern electronic warfare.However,with the continuous use of various new system radars and complex system radars,the electromagnetic environment is extremely deteriorated.Traditional emitter recognition methods have been difficult to meet the requirements of the rapidly changing battlefield environment for recognition performance.In order to solve this problem,this thesis proposes a emitter recognition algorithm based on deep learning.According to the characteristics of radar emitter signal,an optimized convolution neural network is designed.And processed by feature fusion at the decision level,which improves the recognition effect greatly and also enhances the robustness of the recognition system.The main work of this thesis is as follows:Starting from the radar reconnaissance system,the reasons for signal preprocessing are analyzed.The methods of signal denoising,signal normalization,intra-pulse modulation recognition,multi-path signal detection and suppression are deeply studied,which lay a foundation for the extraction of emitter signal characteristics.Aiming at the individual feature extraction of emitter,the extraction methods of time domain envelope feature,frequency domain feature,slice feature of ambiguity function and cyclic spectrum feature of emitter signal are studied and analyzed,which provide stable and reliable classification features for emitter signal recognition.By analyzing the network structure and advantages of convolution neural network,a convolution neural network with three convolution layers,three pooling layers and two full connection layers is designed according to the characteristics of less actual radar data samples.Experiments with measured radar data show that the network can extract more detailed discriminative features of emitter signals.In order to prevent the over-fitting phenomenon of the convolutional neural network designed in this paper,it is improved,and data enhancement,random inactivation strategy and multi-task learning strategy are introduced.Experiments on measured radar data show that these three strategies can effectively prevent over-fitting and improve the recognition rate of radar emitter signals.A multi-feature fusion algorithm based on decision level is proposed,which fuses the output of multi-feature through convolutional neural network,so that the feature information can complement each other and the fault-tolerant rate of the system is improved.The experimental results of real radar data show that the recognition rate of fusion is higher than that of any single feature,and it has strong robustness.