Seismic First Break Picking Using Deep Graph Learning for Improved Accuracy and Stability

Core Concepts

A novel deep graph learning-based approach called DGL-FB that leverages higher-dimensional seismic data to achieve superior accuracy and stability in first break picking compared to 2D benchmark methods.

Abstract

The paper proposes a deep graph learning-based framework called DGL-FB for seismic first break (FB) picking. The key aspects of the method are:

Graph Data Preparation:

Each seismic trace is represented as a node in a large graph.
Edges are established between nodes based on the similarity of their midpoints, connecting traces with high correlation.
To manage the large graph size, a subgraph sampling technique is developed to streamline model training and inference.

Deep Graph Encoder:

A deep graph encoder is designed using the GraphSAGE method to encode the subgraphs into global features.
The encoder leverages a combination of SAGEConv layers with LSTM aggregators and TanH activation to capture the comprehensive information in the subgraphs.

ResUNet-1D Picking Header:

The global features from the graph encoder are concatenated with the local trace signals to form the combined features.
A 1D ResUNet architecture is employed to segment the combined features and detect the first breaks.
A weighted binary cross-entropy loss function is used to supervise the output segmentation, balancing the main loss and auxiliary loss.

The proposed DGL-FB framework is evaluated on a field dataset and compared to a 2D U-Net-based benchmark method. The results demonstrate that DGL-FB achieves a 7.2% higher accuracy and a 99.3% reduction in root mean square error (RMSE) compared to the benchmark, showcasing superior accuracy and stability in first break picking.

Stats

The research is supported by the National Natural Science Foundation of China under grant 12371512.

Quotes

"To model seismic data in higher dimensions, we introduce a novel modeling approach to analyze seismic data on a graph. Each seismic trace is treated as a node, and edges are established between nodes based on the near midpoints of corresponding sources and receivers."
"Consequently, graph analysis provides a more generalized methodology. The application of graph neural network theory in geophysics has gained traction."
"To address this issue, we propose a novel picking framework using the deep graph learning technique. Specifically, we first build a huge graph for the whole survey. Second, to boost the training and inference processes, the subgraphs are sampled from the huge graph. Third, a graph neural network is built to encode the global information of the subgraph. Finally, the combined information of global information and local information is fed to a ResUNet to obtain the FB."

Key Insights Distilled From

Seismic First Break Picking in a Higher Dimension Using Deep Graph Learning

by Hongtao Wang... at arxiv.org 04-15-2024

https://arxiv.org/pdf/2404.08408.pdf

Deeper Inquiries

How can the proposed DGL-FB framework be extended to handle other seismic data processing tasks beyond first break picking, such as seismic event classification or seismic attribute regression?

The DGL-FB framework can be extended to handle other seismic data processing tasks by adapting the graph neural network architecture to suit the specific requirements of the task at hand. For seismic event classification, the graph structure can be modified to capture the relationships between different seismic events, with nodes representing events and edges indicating similarities or correlations. The deep graph encoder can then be trained to extract features that are relevant for event classification. Similarly, for seismic attribute regression, the graph can be designed to incorporate attributes of interest as node features, and the encoder can learn to map these attributes to the desired regression output.

What are the potential limitations of the subgraph sampling technique used in DGL-FB, and how could it be further improved to better capture the global structure of the seismic data graph?

One potential limitation of the subgraph sampling technique in DGL-FB is that it may not fully capture the global structure of the seismic data graph, especially if the sampling is not representative of the entire graph. To address this limitation, the sampling strategy could be improved by incorporating more sophisticated sampling methods, such as random walk sampling or stratified sampling based on node importance. Additionally, adaptive sampling techniques that dynamically adjust the sampling strategy based on the graph structure could be implemented to ensure a more comprehensive representation of the global graph.

Given the success of DGL-FB in leveraging higher-dimensional seismic data, how could the framework be adapted to incorporate additional data modalities, such as well log data or geological information, to further enhance the accuracy and robustness of first break picking?

To incorporate additional data modalities like well log data or geological information into the DGL-FB framework, the graph structure can be expanded to include nodes representing different data sources or modalities. For example, well log data can be represented as nodes connected to seismic traces, capturing the spatial relationships between well logs and seismic data. The deep graph encoder can then be modified to handle multiple types of node features, extracting relevant information from both seismic and additional data modalities. By integrating diverse data sources into the graph structure, DGL-FB can leverage the complementary information to improve the accuracy and robustness of first break picking.