Electromagnetic localization of dielectric targets in a 3D geometry by means of a neural network approach

This paper deals with an electromagnetic inverse scattering problem, i.e. the localization of a dielectric target in a 3D half-space. The approach chosen to face the problem makes use of a learning-by-examples procedure based on a neural network algorithm. The neural network is trained to reconstruct the coordinates of the barycenter of the object starting from the acquisition of the back-scattered electromagnetic field at a number of observation positions. The performances of the approach are analyzed by considering different types of input data and different neural networks setups