Small Sample Identification for Specific Emitter Based on Adversarial Embedded Networks

Abstract

The technology of specific emitter identification (SEI) has important military significance in electronic warfare. However, it is hard to obtain sufficient signal samples from the specific emitter in the electromagnetic environment of the battlefield. Therefore, it is a challenging issue to learn from a handful of samples to accurately identify complex and changeable emitter. A small sample identification method based on adversarial embedded networks is proposed to solve this problem. This method combines the improved generative adversarial networks (GAN) and the Convolutional neural networks (CNN) for classification. In the context of a handful of samples, high-quality simulation samples are generated by the generator in the generative adversarial networks to expand the available feature quantities of the model, thereby improving the recognition efficiency and accuracy. Through the training and testing of a small number of radar emitter data and communication station emitter data, the results show that the method just needs a small amount of data to achieve higher recognition accuracy.