Subsurface Feature Mapping With Masked ML Structural Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying subsurface features struggle with handling large volumes of geophysical data and often require subjective analysis or generate binary results without providing detailed information on structural features.
Innovation Solution
A computer-implemented method using machine learning techniques, such as object identification and mask-RCNN, to train a model that generates subsurface feature data, including structural identification and categorization values, to accurately identify and categorize subsurface features as a function of position.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional methods are used to identify subsurface features, then the process can be completed with simpler tools, but the ability to handle large volumes of geophysical data and provide detailed structural information is insufficient
Solution Approach 1:
The patent replaces traditional mechanical/manual analysis methods with machine learning algorithms and automated computational systems. The system uses trained models to automatically process geophysical data, extract features, and generate structural interpretations, substituting human expert analysis with algorithm-based processing that can handle large data volumes while providing detailed structural information.
Solution Approach 2:
The patent introduces machine learning models as intermediary systems between raw geophysical data and final structural interpretations. These models serve as mediators that process the data through multiple stages (feature extraction, segmentation, classification) to transform complex geophysical signals into meaningful structural information about subsurface features.
2Productivity
If manual analysis methods are used, then the system complexity is lower, but the productivity and ability to process large volumes of data efficiently is reduced
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models using extensive labeled datasets before deployment. The models undergo supervised training phases where they learn to recognize subsurface features from annotated examples, and may undergo fine-tuning with domain-specific data. This preliminary training enables the system to rapidly process new geophysical data without requiring manual analysis for each new dataset.
Solution Approach 2:
The patent replaces manual analytical processes with automated machine learning pipelines that can process large volumes of geophysical data efficiently. The system uses computational algorithms to perform feature extraction, segmentation, and classification operations that would be time-consuming if performed manually, thereby significantly increasing productivity while managing system complexity through modular architecture.
3Measurement precision
If binary classification methods are used, then the analysis process is simpler, but the ability to provide detailed categorization and structural identification is limited
Solution Approach 1:
The patent applies segmentation by dividing the classification task into multiple distinct stages and components. The system segments the analysis into feature extraction, object detection, segmentation, and classification phases, each handled by specialized sub-models. This allows the system to provide detailed structural identification and categorization (beyond simple binary classification) while managing complexity through modular, specialized components rather than a single monolithic model.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Systems and methods are disclosed for identifying subsurface features as a function of position in a subsurface volume of interest. Exemplary implementations may include obtaining target subsurface data; obtaining a conditioned subsurface feature model; applying the conditioned subsurface feature model to the target subsurface data, which may include generating convoluted target subsurface data by convoluting the target subsurface data; generating target subsurface feature map layers by applying filters to the convoluted target subsurface data; detecting potential target subsurface features in the target subsurface feature map layers; masking the target subsurface features; and estimating target subsurface feature data by linking the masked subsurface features to the target subsurface feature data.