Implicit Seismic Full Waveform Inversion With Deep Neural Representation

Full waveform inversion (FWI) commonly stands for the state-of-the-art approach imaging subsurface structures and physical parameters, however, its implementation usually faces great challenges, such as building a good initial model to escape from local minima, evaluating uncertainty of results. In this paper, we propose implicit full (IFWI) algorithm using continuously implicitly defined deep neural representations. Compared FWI, which is sensitive model, IFWI benefits increased degrees freedom with learning optimization, thus allowing start random initialization, greatly reduces risk non-uniqueness being trapped in minima. Both theoretical experimental analyses indicates that, given able converge global minimum produce high-resolution image fine structures. addition, analysis can be easily performed by approximating Bayesian inference various approaches, analyzed paper adding dropout neurons. Furthermore, has certain degree robustness strong generalization ability that are exemplified experiments 2D geological models. With proper setup, also well suited multi-scale joint geophysical inversion.