Deep pre-trained FWI: where supervised learning meets the physics-informed neural networks

An accurate velocity model is essential to make a good seismic image. Conventional methods perform Velocity Model Building (VMB) tasks rely on inverse methods, which, despite being widely used, are ill-posed problems that require intense and specialized human supervision. Convolutional Neural Networks (CNN) have been extensively investigated as an alternative solve the VMB task. Two main approaches were in literature: supervised training Physics-Informed (PINN). Supervised presents some generalization issues since structures, ranges must be similar test set. Some works integrated Full-waveform Inversion (FWI) with CNN, defining problem of PINN framework. In this case, CNN stabilizes inversion, acting like regularizer avoiding local minima-related and, cases, sparing initial model. Our approach combines physics-informed neural networks by using transfer learning start inversion. The pre-trained obtained based reduced simple data set capture trend at FWI iterations. We show reduces uncertainties process, accelerates convergence, improves final scores iterative process.