We propose an algorithm for seismic data interpolation using generative adversarial networks (GANs). The method works by extracting feature vectors of the training data by self-learning and does not require any processing to create the training labels. The algorithm does not make any assumptions about linearity of events or sparsity of the data that are often included in the traditional interpolation methods. We create the training labels by randomly removing traces from different receiver indices of the original data sets to simulate the effect of missing traces. We adopt the framework of CycleGAN algorithm to train the network and add additional loss functions to regularize the model. Numerical examples using land and marine field-data sets demonstrate the validity and effectiveness of the proposed approach. With a small computational burden, the proposed method can achieve accurate interpolation results and can easily be applied to 3D seismic data sets.
Presentation Date: Monday, September 16, 2019
Session Start Time: 1:50 PM
Presentation Time: 1:50 PM
Location: 221D
Presentation Type: Oral
