Geophysical inversion using multilayer perceptron

This paper is a continuation report of the previous research on seabed logging (SBL). In this paper, it was shown that a certain geophysical inverse problem (such as one posed by SBL) can be solved using an important class of artificial neural networks, which is a multilayer perceptron (MLP). To show this, several sets of synthetic data has been generated using some assumed models of a physical property (such as seabed resistivity) distribution. Then, these pairs of data and models were used to train a MLP with a certain architecture. Finally, the trained MLP was tested to do inversion with new data and produced a predicted model. The predicted model was reasonably close to the true model and the mean square error (MSE) between them was 0.016.