With the growing demand of high-resolution subsurface characterization from 3D seismic surveying, the size of 3D seismic data has been dramatically increasing, and correspondingly, the process of interpreting seismic data is becoming more time consuming and labor intensive. Recently, there has been great interest in various supervised machine learning techniques that can help reduce the time and effort consumed by manual interpretation workflows. However, supervised machine learning requires labels for training, and obtaining these labels for large volumes of seismic data is a very demanding and challenging task. In this work, we propose a weakly-supervised approach to labeling subsurface seismic structures using Orthogonal Non-Negative Matrix Factorization (ONMF) with additional sparsity constraints. We show that “rough” image-level labels can be used to obtain high-quality pixel-level labels using weakly-supervised learning. Such pixel-level labels can then be used for training powerful supervised machine learning models. This approach greatly simplifies the process of generating labeled data for applying machine learning techniques to computational seismic interpretation tasks. Results obtained by labeling various subsurface structures from the Netherlands North Sea F3 block show very promising results.
- Y. Alaudah and G. AlRegib, “A weakly-supervised approach to seismic structure labeling,” submitted to 87th Annual SEG Meeting Extended Abstracts, Houston, Texas, 2017. [PDF] [Poster/Lec] [Bib] [Code]
- Y. Alaudah, H. Di, and G. AlRegib, “Weakly Supervised Seismic Structure Labeling via Orthogonal Non-Negative Matrix Factorization”, 79th EAGE Annual Conference & Exhibition, Paris, France, June 12-15, 2017. [PDF] [Poster/Lec] [Bib] [Code]
- Y. Alaudah and G. AlRegib, “Weakly-Supervised Labeling of Seismic Volumes Using Reference Exemplars,” IEEE Intl. Conference on Image Processing (ICIP), Phoenix, Arizona, USA, Sep. 25-28, 2016. [PDF] [Poster/Lec] [Bib] [Code]