Kohonen's neural networks: Application to determine geological models from gravimetric datasets.

@inproceedings{LevassorGM2009,
 abstract = { Processing of geophysical potential fields provides valuable insights on subsurface features, but is computationally demanding. In this paper, we describe how to exploit a catalogue of pre-computed gravity responses on a set of geological bodies in order to speed up the interpretation of gravity anomaly maps. The method is based on a statistical and unsupervised learning approach, called self-organizing map (SOM), also known as Kohonen’s map. SOM allows to reduce a n-dimensional and highly noised signal into two-dimensional maps which can be easily interpreted. This paper summarizes the neural network and the Kohonen’s map theory. The major goal of this study make sure that Kohonen’s networks capabilities for handling gravimetric data sets. Firstly, the methodology is tested on character recognition, then, on recognizing geological models from their respective associated gravimetric data sets. Finally, a model probability approximation is applied to these data sets. This approach decreases considerably the geological model recognition time of uninterpreted gravimetry data sets. },
 author = { Levassor, Fabrice AND Foudil Bey, Nacim AND Caumon, Guillaume },
 booktitle = { Proc. 29th Gocad Meeting, Nancy },
 title = { Kohonen's neural networks: Application to determine geological models from gravimetric datasets. },
 year = { 2009 }
}