Development of a Multi-attribute Convolutional Neural Network to Seismic Facies Classification

Abstract

Seismic facies classification contributes to the interpretation and geological modeling of subsurface.The complexity of geological environments still challenges any standardization of patterns recognition methods for estimating seismic facies and the search for an ideal method, since this step can be complex depending on the purpose and the available dataset.The convolutional neural networks (CNN) were developed to serve as the computer vision, seeking to imitate the functioning of human vision.Despite the potential of this technology, the convolutional networks have not yet been successfully adapted to work with seismic data.The main objective is to develop a workflow based on multi-attribute convolutional neural network for seismic facies characterization.The workflow is based on four main steps: facies classification; data extraction; CNN train and prediction.Aiming to verify the efficiency of this new method, it was elaborated a synthetic image and it was also applied to a real seismic data.In addition, seismic attributes have been calculated to enhance characteristics of the original dataset.The method demonstrates exciting results achieving accuracies above 90% over the expected facies model.