High Resolution ISAR Imaging for Maritime Targets by Encoder-Decoder Network Combined with the Time-Frequency Attention

Abstract

The Inverse Synthetic Aperture Radar (ISAR) imaging result of a rotating maritime target is usually defoculsing since the complex sea conditions may cause time-varying spin speed of targets. A common solution is to utilize range instantaneous Doppler (RID) algorithm for short time imaging, however, the RID algorithm suffers from the low time-frequency resolution. Therefore, this paper proposes an Encoder-Decoder neural net-work combined with Time-Frequency Attention (TFA) module to enhance instantaneous frequency resolution. The TFA module is utilized to extract the spectral lines of weak scatterers, further, the Encoder-Decoder network is designed to transform the low-resolution time-frequency refined by the TFA module to their high-resolution counterparts. Finally, the performance of the proposed method is evaluated on the simulated and real measured data. Both imaging and numerical results certify the meliority of the proposed method in enhancing frequency resolution and reserving the spectral lines of weak scatterers.