Rock Image Constituent Detection With Small-Data ML Counting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting and counting geological constituents in rock samples, such as microfossils, nanofossils, and minerals, are labor-intensive, require extensive human expertise, and involve time-consuming image acquisition and large training datasets, making automation challenging.
Innovation Solution
A method using a location detection machine learning algorithm, preferably a convolutional neural network, to automatically detect and count geological constituents by surrounding them with geometric shapes in a single image, reducing preparation time and operational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual detection and counting by human experts is used, then detection accuracy and classification precision are improved, but productivity is worsened due to tedious and time-consuming processes
Solution Approach 1:
The patent replaces the manual mechanical observation and classification process with an automated image processing system using machine learning algorithms. The system automatically detects, classifies, and counts geological constituents from images, substituting human expert manual work with computational algorithms that maintain accuracy while dramatically improving productivity.
Solution Approach 2:
The system enables self-service automation where the machine learning model independently performs detection, classification, and counting without requiring continuous human intervention. Once trained, the system autonomously processes images and generates results, allowing non-experts to perform tasks that previously required trained specialists.
2Extent of automation
If machine learning methods with large training datasets are used, then automation extent is improved, but loss of time is worsened due to extensive image acquisition and preparation requirements
Solution Approach 1:
The patent applies partial action by using a limited number of training images (at least 10) rather than requiring extensive datasets. This approach achieves sufficient automation performance without the time-consuming collection and preparation of large volumes of training data, balancing automation extent with preparation time investment.
Solution Approach 2:
The system performs preliminary training with a small dataset to establish the machine learning model, which then enables rapid automated processing of subsequent images. This preliminary action creates a reusable model that eliminates the need for extensive per-image preparation, reducing overall time loss while maintaining high automation levels.
3Measurement precision
If multiple images under different conditions are acquired for training, then measurement precision is improved, but device complexity is worsened due to requirements for specialized equipment and multiple acquisition conditions
Solution Approach 1:
The patent creates a universal machine learning model that can classify geological constituents across different image types and conditions using a single trained system. The model is designed to handle various imaging scenarios (different polarizations, lighting conditions) without requiring separate specialized equipment or multiple dedicated systems, reducing device complexity while maintaining precision.
4Reliability
If extensive manual identification and annotation of training images is performed, then machine learning model accuracy is improved, but loss of time is worsened due to consistent expert work requirements
Solution Approach 1:
The patent uses partial action by requiring annotation of only at least 10 training images rather than extensive datasets. This provides sufficient reliability for the machine learning model to achieve accurate classification while dramatically reducing the time investment required for manual annotation compared to traditional approaches that would require hundreds or thousands of annotated images.
Data Source
AI summary
The present invention is a method of detecting and counting a geological constituent (cge) of an acquired image of a rock sample (IER), by a location detection machine learning method (ALG).