Segmentation of 2-D seismic data

Stratigraphic traps containing hydrocarbon deposits are typically difficult to locate using seismic data. For this reason it is estimated that traps of this form contain much of the worlds remaining undiscovered oil and gas. Seismic exploration for these deposits can be augmented by the measurement of subtle attributes and the application of non-linear pattern recognition techniques. In this paper, a simple model of a stratigraphic, hydrocarbon trap is used to create synthetic seismic data. Five features are extracted and examined using histograms of three data classes. Cluster analysis is used to segment the seismogram and to further analyze the discriminatory power of the features. Finally, a non-linear Bayes classifier is applied to the data using two different approximations of the probability density function. The classifier produces 30% wrong classifications when the density function is modeled as Gaussian. Errors are reduced to 8% when the density function is estimated by a multi-modal Gaussian density.