Seismic Facies Segmentation Using Deep Learning

Abstract

Seismic reflection surveying is the most used method to obtain subsurface information in the O&G exploration industry. Through this data, one may obtain structural and stratigraphic geometric features and potential hydrocarbon deposit locations. Even though it is paramount, seismic data interpretation is an extremely time-consuming and human-intensive task, mainly due to the ever-larger volumes of seismic data and the geological complexity present in the study areas. Aiming this issue, computer-aid systems assisting geoscientists to interpret this large and complex data in a faster and more accurate manner represent vital importance for the development of the exploration industry. On the other hand, deep learning techniques are currently applied in several areas of science to support tasks that were considered human-centered, e.g., image classification, language translation, among others. In this work, we created a neural network topology to assist interpreters in the stratigraphic mapping of seismic images at the pixel level resolution.

Our recent results have demonstrated that deep learning can distinguish among different facies helping the interpreter to process new seismic images. We also present a network that can classify parts of the image with high accuracy. Here, we extend this structure to create a neural network that can classify the seismic image at pixel level producing an interpretation mask suggestion. First, we selected a trained convolutional neural network (CNN) with the highest accuracy on the classification task. Then, we modified the final part of the model to produce pixel-wise predictions. Next, we train our neural network using real seismic data from Netherlands Offshore in F3 block. Since these seismic images are somewhat large, we decided to break them into tiles corresponding to 20% of the entire image area. To generate the final prediction, we apply the network throughout the image. In our experiments, we achieved more than 97% of pixel accuracy, and our qualitative results showed that the model could produce a mask very close to the actual interpretation.

AAPG Datapages/Search and Discovery Article #90323 ©2018 AAPG Annual Convention and Exhibition, Salt Lake City, Utah, May 20-23, 2018