Deep Learning: A Powerful Framework for the Real-Time Solution of Inverse Scattering Problems

This work deals with the real-time retrieval of the electromagnetic ( EM ) properties of an inaccessible domain starting from non-invasive measurements. Within this context, deep learning ( DL ) is rapidly emerging as a powerful paradigm to solve inverse scattering ( IS ) and imaging problems on a pixel-basis with unprecedented computational efficiency and accuracy. Accordingly, the main principles and fundamental aspects of the most diffused deep neural network ( DNN ) architectures [i.e., convolutional neural networks ( CNNs )] are briefly summarized, and an overview of the most recent works in this promising field of research is given. Paramount unsolved challenges are finally discussed to indicate some possible lines-of-research to overcome current limitations of DL strategies as applied to IS and imaging.