DNN Application For Pseudo-Spectral FWI

Abstract

Full-waveform inversion (FWI) is a widely used technique in seismic processing to produce high resolution earth models iteratively improved an earth model using a sequence of linearized local inversions to solve a fully non-linear problem. Deep Neural Networks (DNN) are a subset of machine learning algorithms that are efficient in learning non-linear functionals between input and output pairs. The learning process within DNN involves iteratively updating network neuron weights to best approximate input-to-output mapping. There is clearly a similarity between FWI and DNN for optimization applications. I propose casting FWI as a DNN problem and implement a novel approach which learns pseudo-spectral data-driven FWI. I test this methodology by training a DNN on 1D data and then apply this to previously unseen data. Initial results achieved promising levels of accuracy, although not fully reconstructing the model. Future work will investigate deeper DNNs for better generalization and the application to real data.