Fourier-DeepONet: Fourier-enhanced deep operator networks for full waveform inversion with improved accuracy, generalizability, and robustness

Full waveform inversion (FWI) infers the subsurface structure information from seismic data by solving a non-convex optimization problem. Data-driven FWI has been increasingly studied with various neural network architectures to improve accuracy and computational efficiency. Nevertheless, applicability of pre-trained networks is severely restricted potential discrepancies between source function used in field survey one utilized during training. Here, we develop Fourier-enhanced deep operator (Fourier-DeepONet) for generalization sources, including frequencies locations sources. Specifically, employ Fourier as decoder DeepONet, utilize parameters input Fourier-DeepONet, facilitating resolution variable To test three new realistic benchmark datasets (FWI-F, FWI-L, FWI-FL) varying frequencies, locations, or both. Our experiments demonstrate that compared existing data-driven methods, Fourier-DeepONet obtains more accurate predictions structures wide range parameters. Moreover, proposed exhibits superior robustness when handling Gaussian noise missing traces sources noise, paving way reliable imaging across diverse real conditions.