Self-supervised, active learning seismic full-waveform inversion

A novel recursive, self-supervised machine-learning (ML) inversion scheme is developed. It is applied for fast and accurate full-waveform inversion of land seismic data. ML generalization is enhanced by using virtual super gathers (VSGs) of field data for training. These are obtained from midpoint-offset sorting and stacking after applying surface-consistent corrections from the decomposition of the transmitted wavefield. The procedure implements reinforcement learning concepts by adopting an inversion agent to interact with the environment and explore the model space under a data misfit optimization policy. The generated parameter distributions and related forward responses are used as new training samples for supervised learning. The active learning (AL) paradigm is further embedded in the procedure, for which queries on data diversity and uncertainty are used to generate fully informative reduced sets for training. The procedure is recursive. At each cycle, the physics-based inversion is coupled to the ML predictions via penalty terms that promote a long-term data misfit reduction. The resulting self-supervised, AL, physics-driven deep-learning inversion generalizes well with field data. The method is applied to perform full-waveform inversion (FWI) of a complex land seismic data set characterized by transcurrent faulting and related structures. High signal-to-noise VSGs are inverted with a 1.5D Laplace-Fourier FWI scheme. The AL inversion procedure uses a small fraction of data for training while achieving sharper velocity reconstructions and a lower data misfit when compared with previous results. AL FWI is highly generalizable and effective for land seismic velocity model building and for other inversion scenarios.