An innovative Microwave Imaging Approach exploiting the Orthogonality Sampling Method for Physics-guided Deep-Learning

In the last years, there is a growing interest in the integration of deep learning (DL) in microwave imaging (MWI) [1]. As well known, the performances of traditional inversion approaches are tamed by the non-linearity and the illposed of the underlying inverse scattering problem. This often leads to poor results or even false solutions, which are different from the ground truth that cannot be discriminated by the algorithm. The powerful computational tools offered by DL, which offer the capability of retrieving complex non-linear relationships between input and output, would allow to significantly enhance the performance of microwave imaging, possibly overcoming the drawbacks of traditional approaches. Such a perspective is particularly attractive in the field of medical MWI, wherein the availability of reliable, user-independent images is crucial to properly support the clinicians.