Seismic Interpretation Flow Fields for Horizon Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning methods for seismic image segmentation and horizon tracking in subsurface interpretation require large sets of manual labels and are not flexible enough to handle complex geological structures, particularly in tasks like dense horizon extraction and relative geological age volume generation.
Innovation Solution
A method involving the generation of flow fields based on differences between ordered seismic images, using machine learning models to identify objects and track horizons, which reduces the need for extensive manual labeling and enhances the model's ability to interpret complex geological structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If deep convolutional neural networks and machine learning models are used for seismic image segmentation and horizon tracking, then automation extent and productivity are improved, but the need for large training corpora of manual interpretations increases device complexity and loss of time
Solution Approach 1:
The system performs self-training by automatically generating synthetic labeled seismic data from unlabeled seismic volumes. The machine learning model trains on its own generated data without requiring extensive manual annotations, enabling the system to improve its own performance autonomously and reducing the burden of manual label creation
Solution Approach 2:
The system performs preliminary synthesis of seismic data with embedded ground truth labels before actual interpretation tasks. By pre-generating training data with known annotations, the system prepares the necessary training corpus in advance, eliminating the need for time-consuming manual labeling of large datasets
2Measurement precision
If deep learning models are trained with large manual label sets to accurately predict complex geological structures, then measurement precision is improved, but productivity and ease of operation deteriorate due to extensive human intervention
Solution Approach 1:
The system generates its own training data automatically without requiring human annotators to create large labeled datasets. This self-service approach maintains high prediction accuracy by using synthetic data with embedded ground truth, while eliminating the time-consuming manual labeling process and improving overall productivity
Solution Approach 2:
The system creates synthetic copies of seismic data with embedded ground truth labels that replicate the characteristics of manually labeled data. These synthesized copies serve as training samples, providing the model with sufficient learning material without requiring actual manual annotations, thus maintaining precision while improving productivity
3Ease of operation
If traditional machine learning methods are used for seismic interpretation, then ease of operation is maintained, but adaptability to complex geological structures and flexibility for dense horizon extraction are limited
Solution Approach 1:
The system dynamically adapts to complex geological structures by using flow field representations that can capture arbitrary deformations and transformations. The model learns dynamic patterns in seismic data through self-trained flow fields, enabling it to handle diverse and complex geological formations while maintaining ease of operation through automated processing
Solution Approach 2:
The system changes the parameter representation from direct seismic amplitudes to flow field transformations between adjacent seismic images. This parameter transformation enables the model to capture complex structural relationships and deformations, significantly improving adaptability to various geological structures while keeping the operational interface simple
Data Source
AI summary
A method for modeling a subsurface volume includes receiving a plurality of ordered seismic images including representations of objects in the subsurface volume, generating flow fields based on a difference between individual images of the plurality of ordered seismic images, and identifying the objects in the seismic images based on the flow fields and the plurality of ordered seismic images.