Few Samples Radar Signal Recognition based on Conditional Encoding Generative Adversarial Network Model

Deep learning techniques have demonstrated exceptional performance in problems involving the recognition of radar modulation signals. Nevertheless, the amount of radar signals gathered is not enough to match deep neural network training needs given the complex and dynamic electromagnetic settings of current electronic warfare. This study offers a data augmentation strategy based on Conditional Encoding Generative Adversarial Network (CEGAN) to deal with low accuracy of radar signal recognition under few samples. Firstly, we use Encoder to extract the features of time-frequency images (TFIs) and add the feature information to the generation network to improve the quality of generated samples. Next, random conditions are introduced into Conditional Generation Adversarial network (CGAN) to control the categories of generated images. Meanwhile, the objective function combining conditional adversarial loss and L<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> loss is constructed to optimize the network model. Finally, CEGAN generates high quality radar signal TFIs to extended few sample datasets. Simulation results demonstrate that CEGAN can generate more realistic TFIs. And after data augmentation, the overall recognition accuracy of 10 kinds of radar signals has been boosted by 17.70% when the signal-to-noise ratio (SNR) is 0dB.