A Transceiver-Configuration-Independent 2-D Electromagnetic Full-Wave Inversion Scheme Based on End-to-End Artificial Neural Networks

This communication presents a novel end-to-end artificial neural network (ANN)-based 2-D electromagnetic (EM) full-wave inversion (FWI) scheme in which the transceiver configurations are allowed to be different in training and testing. The equivalent current on the boundary of the numerical computational domain bridges the different scattered field data excited or recorded by different transceivers used in the training and testing. The transformation is accomplished through the following three steps: first, the fictitious equivalent current is solved from the scattered field recorded in the testing by multiplying it with the inverse of the radiation matrix whose kennel is the Green’s function of the background medium; then, let the obtained fictitious equivalent current radiate under the configuration of the transceivers adopted during training; finally, the scattered field data recorded by the receivers used in the training are input into the trained ANN to predict the dielectric parameters of the scatterers enclosed in the inversion domain. This rigorous transformation enables the end-to-end ANN training and testing for FWI to be implemented for different transceiver configurations and has many potential applications. Several numerical examples are used to validate the feasibility of the proposed inversion scheme.