Radar signal recognition based on deep convolutional neural network in complex electromagnetic environment

To solve the problem that tradition signal recognition algorithms cannot effectively recognize the contaminated and diverse radar signals in complex and variable Electronic Warfare (EW) environment, a new recognition method based on deep convolutional neural network (CNN) and time-frequency (TF) analysis is proposed. Firstly, the TF images of radar signals are extracted as the inputs to the CNN model. Then, a new network, called CNN-TF, is constructed to analyze these time-frequency images and use the robustness of CNN to suppress the noise interference. Thirdly, a complete and diverse signal librai7 is constructed based on the complex EW environment, and the librai7 is used to train and test CNN-TF. Finally, trained CNN-TF will be used for signal recognition. Simulation results show that the proposed algorithm not only improves the performance of signal recognition, but also has excellent anti-noise performance, which makes the proposed algorithm adapt to the complex and variable electronic warfare environment.