Selecting and Combining Attributes to Enhance Detection of Seismic Objects

F-26 SELECTING AND COMBINING ATTRIBUTES TO ENHANCE DETECTION OF SEISMIC OBJECTS Abstract 1 This paper describes recent experiences with the seismic object detection method developed by Meldahl et al. (1998 and 1999). In this patent pending method supervised or unsupervised neural networks are used to transform multiple ‘directive’ attributes into ‘object probability’ classes. The method is used a/o to detect seismic chimneys and faults (Heggland et al. 1999 and 2000). Selection of attributes is a crucial step in the procedure especially in the unsupervised mode. In this paper we discuss methods and criteria to optimize the attribute selection process. Furthermore