Edge-guided Based Full Waveform Inversion Algorithm

Seismic full-waveform inversion has shown to be able to determine high-resolution velocity model of the reservoir with a spatial resolution on the order of the seismic wavelength and thus to better characterize the model discontinuities or model edges.Compared with the travel-time tomography method,FWI seeks a better velocity model by matching the recorded seismic waveform with synthetic waveform,in phase and/or amplitude.It is well known that FWI is a highly non-linear inverse problem and the solution could be unstable.For regularizing the inversion system,quadratic regularization such as Tikhonov regularization is computationally easy to be applied.However,this kind of regularization applies homogeneous smoothing to the inverted model and,consequently,it will tend to blur sharp model boundaries,or model edges.In comparison,non-quadratic regularization such as total variation can better recover model edges by enforcing sparsity on the model gradients.TV regularization has been proposed and successfully applied to better determine subsurface model edges in seismic imaging.For oil/gas or geothermal reservoir imaging,accurately characterizing subsurface stratigraphic structures and edges is extremely important because they may provide paths for fluids or confine the reservoir boundaries.For subsurface exploration,great efforts have been made on the development and application of various discontinuity detection algorithms.we propose a new edge-guided FWI method that model edges are extracted directly from the inverted models during iterations of FWI.They are then used for guiding the calculation of FWI gradient in a way different from Ma et al.,2012 and at the same time guiding the edge-preserving TV regularization following Lin et al.,2013.Therefore,our proposed FWI strategy avoids the computational burden of seismic migration but keeps the advantages of edge guided calculation of gradient and TV regularization.In order to extract robust edge information from the model,we adopt Canny edge detection algorithm that is widely used in image processing community.We also apply bilateral filtering to remove some'noise' in the FWI gradient to make the inversion more stable.The new FWI method is validated using the complex Marmousi model that contains several steeply dipping fault zones and hundreds of horizons.Compared to FWI without edge guidance,our proposed edge-guided FWI recovers velocity model anomalies and edges much better.