Automatic Geobody Detection Using Multi-class Sparse Representation
G. Larrazabal, C. Ramirez and G. Gonzalez
Event name: 76th EAGE Conference and Exhibition 2014
Session: Seismic Attributes - From Theory to Applications
Publication date: 16 June 2014
Info: Extended abstract, PDF ( 1Mb )
Price: € 20
Salt segmentation and characterization can be viewed as a classification problem where two groups or classes (SALT and SEDIMENT) are to be assigned to each seismic data element at a given coordinate. In a previous work we propose to construct a classification machinery based on the theory of sparse representation in order to carry out salt characterization in an automatic fashion. Nonetheless, the complexity of the seismic data as well as the limitations of the migration algorithms makes this dual distinction a difficult challenge. In order to overcome this obstacle, we propose to extend our original work from the characterization of two classes to a multiclass segmentation framework. In this manner, we allow the sparse representation method to extend the number of possible outcomes when classifying a given test sample thus reducing the number of misclassifications.