Transfer Learning Enhanced Full Waveform Inversion

Full Waveform Inversion (FWI) is a well-known tool in In this contribution we will demonstrate that using the non-destructive testing used to detect the exact location and neural network as discretization of the inverse field, as presented shapes of flaws in a given artefact. To this end, FWI solves in [29], offers the additional advantage of introducing the following inverse-problem: Given a signal emitted at the transfer learning to speed up convergence. This combination boundary of a domain of interest as well as its measured image can outperform classical inversion frameworks in terms of at selected locations, find the material distribution within computational effort by reducing the number of iterations the domain. The solution is found iteratively by computing while maintaining the same quality of reconstruction. The the corresponding forward problem: given a domain with basic idea of this combination is to pretrain the neural a material distribution and signal emitted at the boundary, network on a labeled data set in an offline phase. It is then compute the image of the emitted signal at selected locations.