Data Science Applications in Geophysical Interpretation

Introduction:

Machine learning approaches to geophysical interpretation have recently shown success in automating the identification of structures in seismic data, which in turn can help reduce the time required for the interpretation life-cycle.

I have started to contribute to Landmark’s research team in areas of image segmentation, feature extraction, pattern recognition in salt body detection, fault plane prediction, and multi-facies classification. My experience also includes seismic processing, seismic modeling, inversion, and migration. I have been granted multiple U.S. patents and have several U.S. pending patents. During my research in academia and industry, I have published over 30 peer-reviewed technical papers and gave presentations at different geophysical conferences. Please check out other pages for more detailed projects, publications, and patents.

Below are my recent research and related papers. Some papers are still In Press and will be published soon:

Technology Spotlight 1: Optimize Seismic Interpretation with AI & Cloud Technologies

Technology Spotlight 2: Prediction of Salt Boundaries using Deep Learning Neural Network

AI ML Webinar Recording: Optimize Seismic Interpretation with AI & Cloud Technologies, by Phill Norlund and Fan Jiang, Halliburton Landmark

Leveraging Visual Prompting to Fine-Tune the Segment Anything Model for Seismic Facies Analysis, Jiang et al., [Accepted by 2025 EAGE]

The architecture for fine-tuning Segment Anything Model (SAM) using multi-label visual prompts.
Multiple validation tests of the original SAM and our fine-tuned model. Each color field represents a desired facies type. The IoU scores are significantly improved by fine-tuning SAM with seismic facies data

Implementation of frequency-dependent fault identification by convolutional neural networks with uncertainty analysis, Jiang et al., [PDF]

The workflow to implement frequency-dependent neural network and uncertainty analysis to a fault interpretation process

Improving fault resolution from multiple angle stacks by latent feature analysis with deep learning, Jiang and Osypov, 2024 [PDF]

Implemented a deep learning network to analyze the detailed faults and to extract deep features from latent space (shown at the above picture). The weight coefficient from the original VGG19 network is applied to each angle stack dependent fault volume at different layers.

Implementation of Denosing Diffusion Probability Model for Seismic Interpretation, Jiang and Osypov, IMAGE, 2023 [PDF]

An architecture of the Diffusion Model for seismic interpretation. Seismic image is served as prior to guide the diffusion process.
Diffusion Model inference process from t=T to t=0. (Top) salt segmentation; (bottom) fault prediction.

Implementation of neural style transfer to mitigate domain discrepancy in the seismic salt interpretation, Jiang and Osypov, 2023 [PDF]


The comparisons of salt prediction between different train and test strategies.   

Uncertainty analysis for seismic salt interpretation by deep learning, Jiang et al., 2022, [PDF]

Machine Learning Based Feature Importance Analysis Of Seismic Attributes To Assist Fault Prediction, Jiang and Norlund, 2022, [PDF]

Seismic attributes towards the predictability of seismic features.

* Assisted fault identification and surface extraction by machine learning, a case study from Oman, Jiang et al., 2021, [PDF]

Attribute-assisted fault prediction by deep learning and transfer learning, section & map view (Data courtesy of OXY)
Colored dashed lines represent manual interpretation, black solid lines represent ML prediction.
Fault extraction from the ML prediction volume (Data courtesy of OXY)

* Assisted Fault Interpretation by Multi-scale Dilated Convolutional Neural Network, Jiang and Norlund, 2021, [PDF]


Aggregated multi-scale receptive fields. Both seismic and discontinuity-along-dip will be used as dual-channel data to a CNN architecture. (Modified after Yu and Koltun, 2016)

* The Evolution of Assisted Fault Interpretation — Part 2 [PDF]

A simplified diagram showing how a U-Net Convolutional Neural Network takes an input seismic volume and classifies it into distinct features (fault & not-fault). Resolution = IL, XL, Z, Channels.
A workflow for augmenting ML fault model training with (a) multiple seismic attributes, and (b) real data. Based on Jiang & Norlund, 2020.

* Analysis of Seismic Attributes to Assist in the Classification of Salt by Multi-channel Convolutional Neural Networks, Jiang et al., 2020, [PDF]

(Data courtesy of SEAM)

* Super Resolution of Fault Plane Prediction by a Generative Adversarial Network, Jiang and Norlund, 2020, [PDF]

(Data courtesy of the Australian Government)

* Seismic Attribute-Guided Automatic Fault Prediction by Deep Learning, Jiang and Norlund, 2020, [PDF].

(Data courtesy of New Zealand Petroleum and Minerals)

* Seismic Elastic Wave Modeling with an Adaptive Staggered Grid in Tilted Transversely Isotropic Media, Fan Jiang, 2020, [PDF]

* Seismic Feature Extraction by Attribute-assisted Convolutional Neural Networks, Jiang et al., 2020, In press

* Improve the Resolution of Fault Probability Map by Deep Learning Generative Adversarial Network, Jiang and Norlund, 2020, In press.

Deep learning enabled self-driving car

Under “Projects” page, you will find several projects I have lead and completed in recent years:

  • Geophysical machine learning projects (will update later)
    • multi-label salt body recognition and classification
    • Automatic fault interpretation
    • Super-resolution of the classified fault probability map

  • Seismic data processing & imaging
    • Adaptive grid for anisotropic wave modeling and inversion
    • Hybrid implementation of wave-equation-migration and reverse time migration in anisotropic structure
    • Microseismic modeling with moment tensor sources/simultaneous sources

Last but not least, LIFE is short but colorful, here is my personal photography website, hope you also enjoy it!

Close Bitnami banner
Bitnami