Full text loading...
-
A Quantitative Comparison of Two Convolutional Neural Network Architectures - Seismic Data Interpolation as Example
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 80th EAGE Conference and Exhibition 2018, Jun 2018, Volume 2018, p.1 - 5
Abstract
Accurate seismic data interpolation is important as insufficient sampling and bad traces in seismic data lead to errors in multiple suppression and imaging artifacts. Inspired by the success of the convolutional neural network in the application of image completion and inpainting, we are motivated to use CNN to interpolate seismic data. In this paper, we compare the performance of two CNN architectures, namely the conventional CNN and the convolutional auto-encoder, in an aim to obtain a quantitative understanding towards how network parameters influence the performance. Through synthetic examples, we observe that the two architectures lead to comparable interpolation performances given similar computational constraints.