Thin-Section Image Labeling for Scalable Geological ML Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The scarcity of large-scale, high-quality labeled benchmark data sets for geological thin-section images hinders the development and assessment of machine learning-based geological image analysis and characterization, particularly in reservoir property prediction and wellbore drilling planning.
Innovation Solution
A method and system utilizing web scraping to collect and classify geological thin-section images, employing machine learning models to generate labeled benchmark datasets, which are then used to develop wellbore drilling plans.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to collect and label geological thin-section images, then data quality and accuracy are improved, but the scale and quantity of the dataset are limited due to high cost and specialized requirements
Solution Approach 1:
The patent uses web scraping to automatically copy and collect thin-section images from multiple online sources, creating a large-scale dataset without manually acquiring physical samples. This allows the dataset scale to expand significantly while maintaining quality through automated collection from existing high-quality sources.
Solution Approach 2:
The system employs automated machine learning models that self-label the collected images without requiring continuous human expert intervention. The models automatically classify and annotate images, enabling the dataset to grow at scale while maintaining consistent quality standards through algorithmic validation.
2Reliability
If specialized instruments and skilled operators are used to acquire geological images, then image quality and reliability are improved, but the complexity and cost of the system increase
Solution Approach 1:
The patent replaces physical microscopy instruments and manual image acquisition systems with automated web scraping and digital image collection. This substitution eliminates the need for specialized physical equipment while maintaining image reliability through automated quality validation and consistent digital capture methods.
Solution Approach 2:
The system introduces machine learning models as intermediaries between raw image collection and final labeled dataset production. These models automatically validate, classify, and annotate images, reducing the need for skilled human operators while maintaining high reliability through algorithmic consistency and automated quality checks.
3Measurement precision
If manual labeling by subject matter experts is performed, then labeling accuracy is improved, but the time and resources required increase significantly
Solution Approach 1:
The patent implements self-labeling through machine learning models that automatically analyze and annotate thin-section images without human intervention. The models are trained to recognize geological features and classify images accurately, enabling rapid labeling at scale while maintaining high accuracy through algorithmic validation and confidence scoring.
Solution Approach 2:
The system performs preliminary automated labeling using machine learning models before any human review. This preliminary action pre-processes and pre-labels the entire dataset, allowing human experts to focus only on verifying and correcting ambiguous cases rather than manually labeling every image from scratch, thus dramatically reducing total labeling time.
Data Source
AI summary
A method and a system for generating a labeled benchmark dataset are disclosed. The method includes obtaining a plurality of sources related to a thin section using web scraping and extracting a plurality of images from the plurality of sources related to the thin section, the plurality of images including a plurality of thin section images and a plurality of non-thin section images. Further, the method includes determining the plurality of thin section images from the plurality of extracted images and generating a classification of the plurality of thin section images based on a given classification criteria. The geological thin-section based machine learning models is trained based on the generated classification of the plurality of thin section images and a wellbore drilling plan is generated based on the geological thin-section based machine learning models.