Mineralogy Image Segmentation for Touching Grain Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical petrography and mineralogy methods for analyzing subsurface oil and gas reservoir samples and mine samples are labor-intensive, costly, and challenging to scale, particularly for complex samples, and traditional histogram-based segmentation methods fail to accurately separate touching particles.
Innovation Solution
Employing machine-learning-based filters and cascade classifiers for image segmentation and classification, combined with multimodal imaging techniques to enhance the resolution of EDS images using BSE, SE, XRM, and EM images, and utilizing unsupervised and supervised clustering algorithms to identify and classify grains in light microscopy images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual point counting by skilled petrographers is used, then measurement precision is maintained, but productivity is extremely low and labor costs are high
Solution Approach 1:
The patent replaces manual mechanical point counting by petrographers with an automated image analysis system using machine learning algorithms. The system processes light microscopy images through clustering algorithms (k-means, Gaussian mixture models) to automatically identify and classify mineral grains, eliminating manual labor while maintaining classification accuracy through supervised learning trained on expert-labeled data.
Solution Approach 2:
The system enables self-service automation where the computer vision algorithm independently performs grain identification, segmentation, and mineral classification without human intervention. The machine learning model processes images autonomously, making the system self-sufficient for routine mineralogical analysis tasks that previously required skilled operators.
2Device complexity
If traditional histogram-based segmentation methods are used, then device complexity is low, but measurement precision deteriorates for complex samples with touching particles
Solution Approach 1:
The patent replaces simple histogram-based thresholding with advanced machine learning-based image segmentation. The system uses clustering algorithms that analyze pixel intensity distributions in multidimensional space, enabling accurate separation of touching particles by identifying distinct clusters corresponding to different mineral phases, even when particles are in contact.
Solution Approach 2:
The patent transitions from one-dimensional histogram thresholding to multidimensional clustering analysis. By examining pixel intensities across multiple dimensions and using algorithms like k-means and Gaussian mixture models, the system can distinguish between touching particles of different minerals that would appear overlapping in simple intensity histograms, thereby improving segmentation precision for complex samples.
3Measurement precision
If expensive analytical tools like EDS, XRF, or electron microscopes are used, then measurement precision is improved, but productivity decreases due to scaling challenges
Solution Approach 1:
The patent creates a universal image analysis platform that can process light microscopy images to extract multiple mineralogical parameters simultaneously (grain size distribution, mineral composition, porosity, permeability). This multi-functional system replaces the need for multiple specialized analytical tools, enabling comprehensive mineralogical characterization through a single automated imaging approach that can scale to entire rock cores.
Solution Approach 2:
The system uses light microscopy images as a proxy or copy that contains sufficient information to infer mineral composition and properties without requiring direct chemical analysis. By training machine learning models to recognize mineral-specific optical characteristics in images, the system creates a virtual representation of mineral composition that correlates with actual chemical composition, enabling scalable analysis without expensive analytical instrumentation.
Data Source
AI summary
A method of enhancing a resolution of an EDS image of a sample includes generating an EDS image of the sample, generating a non-EDS image of the sample generating, using a machine learning algorithm, an enhanced resolution EDS image of the sample based on the generated feature map and based on the first EDS image, where a resolution of the enhanced resolution EDS image is higher than a resolution of the first EDS image.