We propose a supervised deep convolutional neural network (CNN) to automatically and accurately characterize paleokarst features in 3D seismic images. To avoid the time-consuming and subjective manual labelling for training the deep CNN, we propose an efficient workflow to automatically generate numerous 3D training data pairs including synthetic seismic images and the corresponding label images of the paleokarst features. With this workflow, we are able to simulate realistic and diverse geologic structure patterns and paleokarst features in the training datasets from which the CNN can effectively learn to recognize the paleokarst features in field seismic images which are not included in the training datasets. Two field examples in the Fort Worth Basin demonstrate that our CNNbased method is significantly superior to the conventional automatic methods in delineating paleokarst features from seismic images and yielding a clear 3D view of all the paleokarst systems from which the geometric parameters of each paleokarst can be automatically and quantitatively measured.
Presentation Date: Monday, October 12, 2020
Session Start Time: 1:50 PM
Presentation Time: 3:30 PM
Location: 351F
Presentation Type: Oral
