An electromagnetic approach based on neural networks for the GPR investigation of buried cylinders

In this letter, neural networks (NNs) are used to reconstruct the geometric and dielectric characteristics of buried cylinders. The NN is designed to work with input data extracted from the transient electric fields scattered by the target. To this aim, a simple simulation of a typical ground-penetrating radar setting is performed and different sets of data examined. Moreover, different neural network algorithms have been exploited, and results have been compared. Finally, the "robustness" of the proposed approach has been tested against noisy data and against uncertainties in the modelization.