Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mapping The Layers of The Ocean Floor With a Convolutional Neural Network

Guilherme G. D. Fernandes, Vitor S. P. P. Oliveira, João P. I. Astolfo

TL;DR

This work tackles the challenge of mapping ocean floor layers via seismic data by employing two UNet-based neural networks to invert seismic shots into velocity fields. Using DeepWave-generated seismic data, the study evaluates UNet and a modified UNet (UnetMod) with 10-fold cross-validation and the Sørensen–Dice Coefficient as the primary similarity metric, achieving DSC values above $0.70$ and notably steady performance for UnetMod. The results highlight that while both networks can capture simpler geological features, complex structures with folds and faults introduce artifacts and reduce accuracy, signaling a need for larger datasets and potentially physics-informed training. The research demonstrates a data-driven, potentially more efficient pathway for ocean-floor mapping in oil exploration, with clear directions for improving robustness and generalization through more data and physics integration.

Abstract

The mapping of ocean floor layers is a current challenge for the oil industry. Existing solution methods involve mapping through seismic methods and wave inversion, which are complex and computationally expensive. The introduction of artificial neural networks, specifically UNet, to predict velocity models based on seismic shots reflected from the ocean floor shows promise for optimising this process. In this study, two neural network architectures are validated for velocity model inversion and compared in terms of stability metrics such as loss function and similarity coefficient, as well as the differences between predicted and actual models. Indeed, neural networks prove promising as a solution to this challenge, achieving Sørensen-Dice coefficient values above 70%.

Mapping The Layers of The Ocean Floor With a Convolutional Neural Network

TL;DR

This work tackles the challenge of mapping ocean floor layers via seismic data by employing two UNet-based neural networks to invert seismic shots into velocity fields. Using DeepWave-generated seismic data, the study evaluates UNet and a modified UNet (UnetMod) with 10-fold cross-validation and the Sørensen–Dice Coefficient as the primary similarity metric, achieving DSC values above and notably steady performance for UnetMod. The results highlight that while both networks can capture simpler geological features, complex structures with folds and faults introduce artifacts and reduce accuracy, signaling a need for larger datasets and potentially physics-informed training. The research demonstrates a data-driven, potentially more efficient pathway for ocean-floor mapping in oil exploration, with clear directions for improving robustness and generalization through more data and physics integration.

Abstract

The mapping of ocean floor layers is a current challenge for the oil industry. Existing solution methods involve mapping through seismic methods and wave inversion, which are complex and computationally expensive. The introduction of artificial neural networks, specifically UNet, to predict velocity models based on seismic shots reflected from the ocean floor shows promise for optimising this process. In this study, two neural network architectures are validated for velocity model inversion and compared in terms of stability metrics such as loss function and similarity coefficient, as well as the differences between predicted and actual models. Indeed, neural networks prove promising as a solution to this challenge, achieving Sørensen-Dice coefficient values above 70%.

Paper Structure

This paper contains 15 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Theoretical Framework
  3. Artificial Neural Networks
  4. Convolutional Neural Network: UNet
  5. Seismic Velocity Model Inversion Using Deep Learning Techniques
  6. Materials and Methods
  7. Datasets
  8. Model Generator
  9. Shot Generator
  10. Google Colaboratory, TensorFlow, and Supercomputer
  11. Artificial Neural Network Training
  12. Statistical Analysis
  13. Results and Discussion
  14. Conclusion
  15. Acknowledgements

Figures (5)

  • Figure 1: Example of a UNet. Given the complexity of the ANNs used in this study, we present an illustration of their architectures instead.
  • Figure 2: Samples of generated velocity models.
  • Figure 3: Loss curves for the two trained network architectures.
  • Figure 4: Comparison between the network's seismic shot input, the predicted models, actual models, and the difference (subtraction) between the predicted and actual models.
  • Figure 5: Sørensen-Dice coefficient for the trained networks, an indicator of the stability of similarity between actual and predicted models by the networks.