Synthetic Aperture Imaging Using Physically Informed Convolutional Neural Networks

We propose a physically informed neural network (PINN) approach to address the problem of locating scatterers using synthetic aperture radar (SAR). The proposed approach uses the physical properties of wave propagation to aid the neural network in learning to detect and localize scatterers from SAR data. This methodology is especially efficient in superresolution cases, i.e. when the scatterers are located close to each other, at distances smaller than the nominal resolution of the imaging system. Using the proposed approach, we are able to detect an a priorly unknown number of scatterers and localize them with high accuracy.