Identification of seismic stratigraphic traps using statistical pattern recognition

Abstract

Stratigraphic hydrocarbon traps commonly result in very subtle changes in seismic reflection waveforms, making their detection difficult using ordinary processing techniques. This paper describes the implementation of a Bayes classifier using a multimodal estimate of the conditional class probability density function. Also, a relaxation labeling procedure is presented which is used to reclassify the initial Bayes results using a priori contextual information extracted from the training data. Processing synthetic data with 20% additive noise resulted in 84.8% correct using only 3 features.