Complete sequence stratigraphy from seismic optical flow without human labeling
Zhun Li, Aria Abubakar
Abstract
Sequence stratigraphy is important to determine geological time-space distribution of strata. A complete sequence stratigraphy interpretation follows the principles of relative geological age determination. For example, some basic principles include principle of original horizontality, principle of lateral continuity and principle of superposition. In another word, any horizon tracked from the stratigraphy volume should be younger than a horizon below, and all tracked horizons should meet loop-tie criterion. We developed a FlowNet based deep learning workflow to generate optical flow cube for a seismic survey. Based on the optical flow, a complete stratigraphic sequence volume can be extracted and obey the basic principles for relative geological time. training data are generated by applying random elastic deformation to original seismic images without human labeling. Flow field is set up to include only vertical dimension to obey the principle of horizontality. Vertical order constraint was applied to both training data and prediction results to ensure that the principle of superposition is followed. Optical flow is cross calibrated between forward and backward direction at both training and prediction phases to enhance the accuracy of flow prediction. Relative geological time volume can finally be generated from optical flow field. Presentation Date: Wednesday, October 14, 2020 Session Start Time: 9:20 AM Presentation Time: 10:35 AM Location: Poster Station 10 Presentation Type: Poster