1-D Inversion of GREATEM Data by Supervised Descent Learning

Abstract

In this letter, the application of the supervised descent method (SDM) for solving controlled-source electromagnetic inversion is studied. The descent direction in each iteration step of the 1-D full-wave inversion (FWI) is learned from the training data set with certain prior information in the off-line training and then saved. In the online prediction, it is directly combined with the measured data and the forward model to implement the FWI. Compared with the traditional iterative method, the efficiency is significantly enhanced since the computation of the Jacobian matrix is circumvented. Both the synthesized and field-measured grounded electrical-source airborne transient electromagnetic (GREATEM) data are used to verify the feasibility and efficiency of SDM. In addition, the learning ability of the SDM is also studied.