Analysis of U-NET Super-Resolving Capabilities in Radar Imaging

This paper investigates the super-resolving power of UNET when applied to a radar imaging problem. In particular, we consider a 2D free-space scenario probed by a multimonostatic/ multifrequency measurement configuration and formulate the imaging problem as a linear inverse scattering one. In this context, we analyze how the degree of ill-posedness of the inverse problem affects the resolution limits achievable by the U-NET. To this aim, the system point spread function, i.e. the reconstruction of a pointlike target, is evaluated via the classical truncated singular value decomposition and used as input to the network. Numerical results relevant to different configuration parameters confirm the “superresolving” capability of the network, but the achievable performance turns out to be strongly dependent on the amount of information in the scattered field data