Full text loading...
-
Multi-Domain Diffraction Identification Using Deep Learning
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 81st EAGE Conference and Exhibition 2019, Jun 2019, Volume 2019, p.1 - 5
Abstract
Seismic imaging is an important aspect for hydrocarbon exploration. A conventional seismic workflow aims to enhance the specular reflections in lieu of the rest of the wavefield. The reflections however, can be limited in imaging sharp corners, for example those seen in fault zones and pinch-outs. Diffractions, on the other hand, form by the interaction of the wavefield with an object which is small in comparison to the wavelength, making them ideal for imaging sharp corners. However, as diffractions have amplitudes an order of magnitude lower than reflections, they must be carefully isolated and processed separately. Here we propose a novel multi-domain deep learning method for separation.
Deep learning is a strand of machine learning which involves a layered neural network model. Here we have manually classified diffractions, reflections, and noise, as well as diffractions which underly reflections, from a range of synthetic and real data using various domains. This classified data is then used to train a convolutional neural network. As more data is input into the neural network, it can adapt and improve, giving better separation. When compared with existing methods, multi-domain deep learning allows for accurate separation of diffractions and reflections as well as removing unwanted noise.