Using Interactive Visualization and Machine Learning for Seismic Interpretation

This book chapter describes the use of interactive visualization and artificial intelligence (AI) for seismic interpretation purposes. After an introduction with some basics about finding oil and gas through seismic interpretation in Sect. 1, we describe two interactive visualization methods called user-driven seismic volume classification in Sect. 2 and semi-automatic detection of anomalies in seismic data based on local histogram analysis in Sect. 3. In Sects. 4 and 5, we describe our approach to use convolutional neural networks (CNNs), a class of deep neural networks, for the detection of geobodies such as fault, channels, and salt domes. In seismic interpretation, confidence in the results and risk minimization is always very important. Which decisions can be made on the results of an artificial intelligence? To address this common concern, we describe in Sect. 6 a method how to understand the operation of the CNNs better and how to thereby increase trust in the findings based on artificial intelligence. As Fraunhofer is about applied research, we had to implement a prototype or solution for our AI-based methods. We called it DeepGeo. We describe DeepGeo in Sect. 7 of this book in detail, before we give a short overview on the status quo on the next things to come from us in the seismic interpretation field with artificial intelligence and deep neural networks.