SRCNN-Based Enhanced Imaging for Low Frequency Radar

Abstract

In this paper, the deep learning based single image super-resolution method is utilized to enhance the quality of radar image. First a sparse-coding like 3-layer convolution neural network is used to enhance the radar image, and the relationship between the convolution neural network and the sparse coding based method is discussed. Then, a deeper neural network with 6 layers is used to enhance the radar image. For both of these neural networks, the input is complex radar image, and they'll output the radar cross section distribution image. Both of these neural networks can sharpen the main lobe, suppress the sidelobe and grating lobe, while the deeper neural network has better performance. The feasibility of the proposed method is testified by simulated data.