Research On Radar Emitter Signal Recognition Method Based On Hybrid Neural Network

The recognition of radar emitter signal is an important part of electronic countermeasure reconnaissance,which can help to judge the system,model and purpose of the enemy radar,and provide an important basis for the following electronic countermeasure.The traditional methods of radar emitter signal identification can not meet the requirements of radar emitter signal identification under the new situation because of the continuous renewal of radar system and the increasingly complex electromagnetic environment.The method of radar emitter signal recognition based on depth learning can automatically extract the signal deep features for classification,and realize the intelligent signal recognition.However,the method of radar emitter signal recognition based on depth learning still has some problems,such as too long training time of model,high requirement of algorithm deployment and large demand of model training samples,etc.,therefore,based on the neural network theory,this paper continues to study the application of the depth learning method in the radar emitter signal recognition field,and has done the following work:1.In this paper,conventional signal,frequency modulation signal,phase modulation signal and complex modulation signal are summarized,and 15 kinds of radar in-pulse modulation signal are selected for simulation modeling,the timefrequency diagram is obtained by time-frequency analysis,and the time-frequency data set of emitters is constructed.2.A method of radar emitter signal recognition based on time-frequency analysis and CNN-BLS network is proposed.In this method,the radar emitter signal is first transformed by CWD time-frequency transform to obtain the signal time-frequency graph,and then the CNN network with multi-scale convolution kernel is used to extract the rich features of the signal time-frequency graph,then the features are sent to the BLS network for signal classification and recognition.The experimental results show that the proposed method effectively balances the recognition accuracy and training time,and preserves the advantage of fast training time of BLS network,and improves the recognition accuracy of network model,the overall recognition rate of the signal is98.727%,which is 0.64% higher than that of Res Net,and the training time is reduced by 15% under the same conditions.3.A method of radar emitter signal recognition based on improved Mobile Vi T network is proposed.This method makes use of the advantages of Mobile Vi T module in feature extraction,which not only extracts the local features of signal time-frequency graph,but also obtains the global features of signal time-frequency graph by Transformer operation,at the same time,the network model uses the improved Mobile Net V2 lightweight network module to achieve the lightweight network.The experimental results show that the algorithm achieves the recognition accuracy matching with the depth model under the condition of using less network parameters and computational load,and the overall recognition accuracy of the signals reaches98.35%,suitable for deployment on computing resource-constrained devices.4.A method of radar emitter signal recognition with small sample based on FFBDC network is proposed.The method uses meta-learning strategy to train and uses brown distance covariance as metric feature to realize small sample radar emitter signal recognition.Different from the original Deep BDC network,the proposed method fuses the multi-scale features in the feature extraction stage,which effectively improves the recognition performance of the network.The experimental results show that the average recognition accuracy of the proposed method is 14.24% and 9.39% higher than that of the conventional transfer learning method in the multi-small sample tasks with only one sample and five samples,respectively,it has good recognition performance in the task of small sample radar emitter signal recognition.