Multiscale Data-Driven Seismic Full-Waveform Inversion With Field Data Study

Seismic full-waveform inversion (FWI), which uses iterative methods to estimate high-resolution subsurface models from seismograms, is a powerful imaging technique in exploration geophysics. In recent years, the computational cost of FWI has grown exponentially due increasing size and resolution seismic data. Moreover, it non-convex problem can encounter local minima limited accuracy initial velocity or absence low frequencies measurements. To overcome these issues, we develop multiscale data-driven method based on fully convolutional networks (FCN). preparing training data, first real-time style transform create large set synthetic natural images. We then two neural with encoder-decoder structure reconstruct low- high-frequency components models, separately. validate performance our effectiveness synthesized set, compare conventional physics-based waveform approaches using both field These numerical results demonstrate that, once model trained, significantly reduce computation time, yield more accurate comparison FWI.