Generating stochastic geological cross-sections using formal grammar

@inproceedings{EDELMANN_RM2021,
 abstract = { Image request is increasingly used to query a knowledge base. In structural geology, crosssections are a support of knowledge, allowing geologists to synthesize the information about the architecture of subsurface and its geological history. Querying databases of geological knowledge using cross-section images could be then an interesting way to rapidly obtain information about analogue subsurface domains studies. Deep-learning techniques are widely used in algorithm dedicated to image queries. These required pairs of “normalized” input/output during the learning phase. Inputs are made of raster images and outputs are composed by tagged objects in the image. Outputs may also include bonds between these objects. Those pairs of data are difficult to gather in the case of geological cross-sections, as it requires a lot of time for the experts to interpret those images. Therefore, it is presented in this paper to generate stochastically geological cross-sections, while keeping the information on the geological processes involved in the settings of the geological architecture represented in the associated cross-section. In this way, it is possible to generate pairs of input/output for learning. The suggested approach combines a formal grammar with a stochastic process to generate 3D structural models, contained in a narrowed volume of interest. This approach has been developed for reproducing geological structures stemming from extensive systems. Beyond the interest of generating pairs of input/output for deep-learning, this technique could also be used for stochastic simulation of 3D structural models. },
 author = { Edelmann, Matthieu AND Foucour, Thomas AND Viseur, Sophie AND Popineau, Fabrice AND Henry, Pierre AND Gout, Claude },
 booktitle = { 2021 RING Meeting },
 publisher = { ASGA },
 title = { Generating stochastic geological cross-sections using formal grammar },
 year = { 2021 }
}