Traditional Synthetic Aperture Processing Assisted GAN-Like Network for Multichannel Radar Forward-Looking Superresolution Imaging
Wenchao Li, Ziwen Wang, Rui Chen, Zhongyu Li, Junjie Wu, Jianyu Yang
Abstract
Radar forward-looking imaging has important applications in autonomous landing, autonomous navigation, reconnaissance guidance and other fields. However, conventional single channel synthetic aperture radar (SAR) or Doppler beam sharpening (DBS) technology has a blind area for forward-looking imaging due to left/right ambiguity and small angle variation. Multichannel radar can utilize the differences of echoes from multiple channels in azimuth to resolve left/right ambiguity, and has the potential for forward-looking imaging. However, there is still a problem of low azimuth resolution due to the restriction of array size. In this article, a deep learning based multichannel radar forward-looking super-resolution imaging framework is proposed. In this framework, synthetic aperture processing is conducted on the echo data of each channel to obtain the image with left/right ambiguity, and the preliminary forward-looking imaging is achieved first by resolving left/right ambiguity with multichannel data. Then, the generative adversarial network (GAN)-like network with mixed attention mechanism is designed to learn the mapping relationship between the original scene and the preliminary imaging result. At last, based on the learned mapping relationship, the echo data of multichannel radar is processed with the proposed framework to achieve forward-looking superresolution imaging. Experimental results were provided to verify the effectiveness of this imaging framework.