Neural Networks for Inverse Problems: the Microwave Imaging Case

In the framework of inverse scattering problems, this paper investigates the use of different artificial-neural network architectures for imaging purposes by processing the data collected at receivers locations in a multiview-multistatic fashion. Generally, this type of problems is strongly nonlinear and ill-posed, thus the development of fast and reliable approaches is paramount for practical implementations. In the last years, machine learning approaches have proved to be very promising to recover quantitatively the electromagnetic features of objects located in an investigation domain, but at the expense of large data sets required for the training procedure. More in detail, this communication tries to explore the role of the network topology by exploiting three different scenarios with an increasingly higher degree of non-linearity.