InversionNet: A real-time and accurate full waveform inversion with convolutional neural network

Abstract

With data proliferation in all geosciences domains, machine learning and data analytics are emerging as important research areas in geosciences. Full-waveform inversion has been an important tool to infer the subsurface based on geophysical measurements. However, solving full-waveform inversion can be challenging. The existing computational methods for solving acoustic full-waveform inversion are not only computationally expensive but also yields low-resolution results because of the ill-posedness and cycle skipping issues of full-waveform inversion. To resolve those issues, we employ machine-learning techniques to solve full-waveform inversion in this work. In particular, we build a convolutional neural network to model the correspondence from data to velocity structures. Our numerical examples using synthetic reflection data show that our new methods much improve the accuracy of the velocity inversion. Furthermore, the computational time for inverting the seismic data is significantly reduced.