Abstract: ‍ ‍With the rapid development of deep learning in the field of image enhancement， it has shown superior performance in the field of ultra-wideband multiple-input multiple-output （MIMO） radar image super-resolution. However， most of the existing research is based on radar image to recover ideal image， but the radar imaging process often requires approximation and quantization operations and relies on the assumption of isotropic target scattering cross-sectional area， which leads to information loss and can affect the upper limit of radar image enhancement performance. Moreover， the imaging results obtained by ultra-wideband radar usually have problems such as low resolution， high grating side lobes and serious clutter pollution， which restrict its practical application. In order to solve the above problems， this paper combines the feature learning of convolutional neural networks and the correlation of projection maps to carry out research， and proposes an ultra-wideband radar image enhancement algorithm based on array subchannel projection maps. Firstly， the correlation of the array subchannel echo of the ultra-wideband radar imaging system is used to construct the array subchannel projection and position coding module， and the projection map and position coding map are used as network extension inputs to enhance the radar image， improve the suppression effect of the network on the radar image grid sidelobe， and obtain higher precision radar image. This paper comprehensively describes the end-to-end ultra-wideband radar image enhancement method from the perspectives of ultra-wideband radar imaging principle， radar image enhancement network structure framework， and radar image enhancement method based on array subchannel projection map. The paper verifies the accuracy of the proposed algorithm using simulation and measurement results. Compared with the traditional algorithm， this method can effectively sharpen the main lobe， suppress the radar image grid sidelobe and clutter， improve the radar image quality， and provide higher quality radar image as input for subsequent abstract applications such as radar image detection and recognition and manual identification.