Mohammad H. Taufik, Fu Wang, and Tariq Alkhalifah

GitHub: https://github.com/DeepWave-KAUST/diffefwi

Multi-parameter elastic full waveform inversion (FWI) provides us with more realistic elastic properties at the cost of significantly increased complexity than standard mono-parameter acoustic FWI. In this work, we employ diffusion models as learned regularization to reduce the ill-posedness of the problem with negligible computational overhead. We demonstrate that even by utilizing only the vertical component of particle velocity, and achieve realistic elastic parameter ratios and (unlike conventional regularization) better data fitting.