An LSTM based strategy for data and model calibration in subsurface electromagnetic imaging

Mismatch between measurements and model used inside the inversion scheme represent a critical point in subsurface electromagnetic imaging. In this paper, a timedomain strategy based on the use of long short-term memory cells is proposed. The approach is assessed through a set of simulations where a couple of antennas are used to acquire scattered-field data in quasi-monostatic mode. A half-space numerical environment is considered. The processed data are then provided in input to a nonlinear multi-frequency inversion algorithm operating in the context of nonconstant-exponent Lebesgue spaces. Initial tests show promising outcomes.