Constrained automatic tops picking using convolutional neural networks
Jeff Godwin, John Roberts, Emily Panetti · 2019
Manual formation top correlation across basins is one of the most labor intensive geological interpretation workflows. Convolutional neural networks (CNN) provide one possible approach to automatically correlate formation tops across a large area with a minimal amount of human effort. We investigate the suitability of a convolutional neural network for automated formation top picking, and combine synthesized depth class probabilities with a prior probability distribution which allows us to constrain possible pick locations. The net result is that constraining the problem in this way improves the accuracy of our predicted pick locations, at the cost of reducing the number of total picks made. Overall, the proposed approach is able to rapidly train a CNN based model and then use that model to make predictions on unpicked wells with very high accuracy, and for relatively low computational effort on commodity hardware. Presentation Date: Tuesday, September 17, 2019 Session Start Time: 9:20 AM Presentation Time: 9:20 AM Location: Poster Station 2 Presentation Type: Poster