Tools
Tools
Chandrashekar, Anusha Gorur (2020) A Deep Neural Network Framework for Seismic Image Classification & Analysis. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Configuration manual)
Download (1MB) | Preview |
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Abstract
Seismic imaging is an indispensable tool in oil and natural gas (O&G) exploration as it helps to delineate the earth's subsurface structures and monitoring through the geometry of sources-receivers coordinates. In the contemporary era, it demands large volumes of seismic data be analyzed and interpreted under standard procedures. Therefore, computational framework systems can stimulate the expert in the classification of subsurface structures/facies to accelerate the analysis process which is paramount to the growth of the industry. The rise in popularity of deep learning motivated scientists to extend those methods to 3D seismic data. While this technique manifested positive returns, the complexity of finding a good starting point for optimizing the parameters of the model is a conventional problem in deep learning systems. Poor or random initialization may lead the network to longer training sessions, vanishing gradients due to backpropagating till initial layers and failing to find the solution. To address this issue, the use of transfer learning with state-of-art deep neural network models are utilized to set a good initialization point to the parameters of the neural network models. The data used is F3 Dutch seismic cube data widely available as open-source data. In seismic data, different facies can be identified by differences in the signatures of the amplitudes, which allows us to label different facies, to assist the process nine facies labels of the data is available from Project MalenoV repository. To achieve this, leveraging the transfer learning with the use of pre-trained models VGG-16, ResNet and Efficient B7 architectures to analyze and interpret the seismic facies. Each model is optimized and evaluated using classification and regression metrics such as Precision, Mae, Categorical accuracy. As a result, VGG-16 and Efficient B7 outperformed ResNet with accuracy 97.4% and 99.1%, MAE 0.085, and 0.008 respectively.
Keywords: Seismic Interpretation, Transfer Learning, Convolutional Neural Networks, Oil, and Gas Exploration
| Item Type: | Thesis (Masters) |
|---|---|
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QA Mathematics > Computer software T Technology > T Technology (General) > Information Technology > Computer software |
| Divisions: | School of Computing > Master of Science in Data Analytics |
| Depositing User: | Dan English |
| Date Deposited: | 20 Jan 2021 14:10 |
| Last Modified: | 20 Jan 2021 14:10 |
| URI: | https://norma.ncirl.ie/id/eprint/4394 |
Actions (login required)
 |
View Item |