Research On Radar Emitter Recognition Algorithm Based On Deep Convolution Neural Network

With the electromagnetic environment becomes increasingly complex,new system radar signals are emerging,making the radar signal recognition of traditional parameters unable to meet the needs of modern electronic reconnaissance.Existing in-pulse feature analysis methods are manually extracted features to identify radar signals,and the calculation amount is too large.Therefore,in order to adapt to the diversified electromagnetic environment and solve the problems in radar signal recognition,thesis proposes a radar emitter identification method based on deep convolutional neural network(DCNN).The method can autonomously learn features,effectively combine feature extraction with target recognition,achieve "end-to-end".The specific research contents are as follows:(1)The radar signal parameters are designed to generate radar data sources.For 8 types of complex radar signals: LFM,LFMCW,LFM-BC,Frank-LFM,S-type NLFM,Costas,P3,FSK/PSK signal modeling and time domain and frequency domain waveform simulation and analysis.By changing the parameters of the communication physical meaning in the radar signal model,it provides data basis for the research on the identification algorithm.(2)A single channel radar emitter identification algorithm based on DCNN is proposed.The main idea is to perform short-time Fourier transform on the eight types of radar signals to obtain spectrogram,and then perform data enhancement and image preprocessing.Different network models are designed for the input real radar waveform and the spectrogram.We analyz the parameters such as the number of training times and the learning rate of the DCNN.It is verified that the radar signal has improved the recognition rate and operation efficiency after STFT and image preprocessing.(3)A two-channel radar emitter identification algorithm based on DCNN is proposed.This method optimizes the single-channel network structure and designs a dual-channel network structure.It can input all real and imaginary information into the network,which is different from the DCNN single-channel signals recognition algorithm.The experiment shows that the dual-channel network has certain advantages in recognition efficiency.It is also verified the signal performance rate is positively correlated with the number of samples.Moreover,the recognition efficiency of the algorithm can reach 85% under low signal-to-noise ratio.Compared with SVM and Adaboost algorithm,the recognition rate of dual channel model is higher.