Specific Emitter Identification via Squeeze-and-Excitation Neural Network in Frequency Domain

Abstract

Specific emitter identification (SEI) is a technology that identifies the emitter individual using the characteristic of their transmit signal. SEI has been widely adopted in military and civilian use. As for SEI task, in time domain, general neural network has a poor feature extraction ability that usually gets a low recognition accuracy in a low signal-to-noise ratio (SNR) environment. So, we propose a SEI method based on squeeze-and-excitation neural network (SeNet) in frequency domain. SeNet uses a squeeze-and-excitation module to achieve channel attention mechanism, so the neural network can emphasize the useful feature and weaken the useless feature in different channels. The experiment result shows that, in frequency domain, SeNet has better emitter identification ability and stronger robustness than the general neural network even in a low SNR environment.