3D reflection tomography allows the macro-model of complex geological structures to be reconstructed. In the usual approach, the spatial distribution of the velocity field is discretized by regular grids. This choice simplifies the development of the related software, but introduces two serious drawbacks: various domains of the model may be poorly covered, and a relevant mismatch between the grid and a complex velocity field may occur. So the tomographic inversion becomes unstable, unreliable and necessarily blurred. In this paper we introduce an algorithm to adapt the grid to the available ray paths and to the velocity field in sequence: so we get irregular grids with a locally variable resolution. We can guide the grid fitting procedure interactively, if we are going to introduce some geological a priori information; otherwise, we define a fully automatic approach, which exploits the Delauny triangles and Voronoi polygons.
A Bayesian filtering approach for inclusion detection with ultrasound reflection tomography