A Machine Learning Benchmark for Facies Classification (1901.07659v2)

Abstract: The recent interest in using deep learning for seismic interpretation tasks, such as facies classification, has been facing a significant obstacle, namely the absence of large publicly available annotated datasets for training and testing models. As a result, researchers have often resorted to annotating their own training and testing data. However, different researchers may annotate different classes, or use different train and test splits. In addition, it is common for papers that apply machine learning for facies classification to not contain quantitative results, and rather rely solely on visual inspection of the results. All of these practices have lead to subjective results and have greatly hindered the ability to compare different machine learning models against each other and understand the advantages and disadvantages of each approach. To address these issues, we open-source a fully-annotated 3D geological model of the Netherlands F3 Block. This model is based on the study of the 3D seismic data in addition to 26 well logs, and is grounded on the careful study of the geology of the region. Furthermore, we propose two baseline models for facies classification based on a deconvolution network architecture and make their codes publicly available. Finally, we propose a scheme for evaluating different models on this dataset, and we share the results of our baseline models. In addition to making the dataset and the code publicly available, this work helps advance research in this area by creating an objective benchmark for comparing the results of different machine learning approaches for facies classification.

• The paper introduces a new public 3D geological dataset and benchmark for machine learning facies classification using seismic data and well logs from the Netherlands F3 Block.
• Baseline deep learning models demonstrate that section-based approaches generally outperform patch-based ones and show significant accuracy improvements with data augmentation.
• This open-source dataset and code create a standardized platform for researchers to objectively compare and advance machine learning methods in seismic interpretation tasks.

Overview of "A Machine Learning Benchmark for Facies Classification"

The paper "A Machine Learning Benchmark for Facies Classification" focuses on addressing a major obstacle in seismic interpretation tasks, particularly facies classification, with the use of supervised machine learning models. Specifically, the scarcity of publicly available and fully annotated datasets for training and evaluating these models hinders progress in the field. Researchers often resort to manual annotation, leading to subjective results that complicate comparisons between various approaches. The authors aim to overcome this challenge by open-sourcing a comprehensive 3D geological model of the Netherlands F3 Block, featuring fully annotated seismic data and well logs. This novel dataset serves as a standardized benchmark to objectively compare machine learning models designed for facies classification.

Key Contributions

1. Dataset Creation: The paper introduces a publicly accessible dataset, grounded in the geology of the Netherlands F3 Block. It integrates 3D seismic data with 26 well logs, ensuring high-quality annotations based on the region's geological paper. Additionally, they extract fault planes for research on fault detection, although these are not applied in the current work.
2. Baseline Models: Two deep learning models using a deconvolution network architecture are proposed as baselines for facies classification. The first model is patch-based, trained on small patches from seismic data, and the second model is section-based, trained on entire inline and crossline sections.
3. Evaluation Protocol: The authors outline a scheme for evaluating models on this dataset, employing metrics such as pixel accuracy, class accuracy, mean class accuracy, and frequency-weighted intersection over union.
4. Open Source Code: The code used to train and test these baseline models is made publicly available, facilitating further research and reproducibility in seismic interpretation using machine learning.

Numerical Results and Insights

The baseline models indicate substantial accuracy improvements when data augmentation and skip connections are incorporated. For example, data augmentation increased both pixel accuracy and mean class accuracy significantly—by more than 10% for patch-based models. This suggests that addressing class imbalance and leveraging additional training data are pivotal in refining model performance.

Section-based models generally outperform patch-based models in both computational efficiency and classification accuracy across lithostratigraphic classes, showcasing the advantages of contextual learning when dealing with entire sections.

Theoretical and Practical Implications

Practically, the release of this dataset is expected to be a cornerstone for researchers aiming to develop and validate advanced machine learning approaches for facies classification. It eliminates challenges associated with subjective dataset annotation and promotes an objective comparison across different methodologies. Theoretically, the detailed geological model of the F3 Block offers insights into lithostratigraphic unit characteristics, supporting developments in regional geology modeling and hydrocarbon exploration.

Future Directions

The public availability of this dataset and baseline models sets the stage for diverse explorations in seismic interpretation. Advanced models may investigate leveraging unsupervised or semi-supervised approaches, and explore hybrid architectures that integrate additional geophysical parameters. Such developments could yield comprehensive models that outperform current baselines in accuracy and efficiency, furthering the potential for automated seismic assessment in complex geological settings. Additionally, wider adoption of standardized datasets like this could catalyze collaborative advancements in geophysical machine learning research globally.

In summary, this paper provides a vital resource for advancing seismic interpretation tasks and benchmarks machine learning techniques, offering transparent and unbiased comparison. The implications extend beyond machine learning, potentially guiding geological insights and exploration efforts.