NORMA eResearch @NCI Library

Deep Neural Network for Seismic Image Segmentation and Detection of Salt Domes

Chambial, Nikhil (2022) Deep Neural Network for Seismic Image Segmentation and Detection of Salt Domes. Masters thesis, Dublin, National College of Ireland.

[thumbnail of Master of Science]
PDF (Master of Science)
Download (1MB) | Preview
[thumbnail of Configuration manual]
PDF (Configuration manual)
Download (2MB) | Preview


Historically, seismic images have been essential in understanding sub-surface structures. Salt Domes is a sub-surface structure that traps hydrocarbons. Domain experts such as geologists use these seismic images to detect salt domes and explore oil and natural gas reservoirs. However, domain experts require substantial computational power to analyze these seismic images, and the classification is also error-prone. Moreover, the precise position of the salt deposit is critical since drilling in the wrong spot would result in significant expenses for corporations as well as environmental harm. As a result, computational systems can stimulate experts in categorizing subsurface structures such as salt domes to fasten the analysis process, which is critical to the industry’s growth. Deep learning’s growing popularity prompted academics to apply such technologies to seismic data. While these methods have produced promising results, the difficulty in determining an appropriate initialization for adjusting the model’s parameters is a common issue in deep machine learning systems. Often the random initialization results in longer training sessions, failure to locate the solution, and disappearing gradients till the first layers owing to back-propagation. To solve this problem, modern deep neural network models with transfer learning are used to provide an initialization point for the model’s parameters and improve the performance. Most of the deep learning models in the past showed sub-par performance due to the scarcity of data that has been resolved in this study by using data augmentation techniques such as rotation and flipping. This study has implemented a state-of-the-art encoder-decoder model called Unet to detect salt domes in seismic images. The model has been trained and tested on the publicly available seismic image dataset provided by TGS. The Unet model is evaluated and optimized using various evaluation metrics such as Intersection over Union(IoU) and Accuracy, with values of 0.80 and 97%, respectively.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Salt Domes; Seismic Images; UNet; IoU(Intersection over Union); Data Augmentation; Convolution Neural Network
Subjects: Q Science > QA Mathematics > Electronic computers. Computer science
T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science
Q Science > QE Geology
Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions: School of Computing > Master of Science in Data Analytics
Depositing User: Tamara Malone
Date Deposited: 19 Jan 2023 15:44
Last Modified: 06 Mar 2023 15:43

Actions (login required)

View Item View Item