Rock Classification via Image Descriptors and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rock classification methods are error-prone and imprecise due to their reliance on direct supervised classification from rock images without considering naturalistic features.
Innovation Solution
A rock classification method using a decision tree and machine learning to classify rocks based on naturalistic descriptors from a rock image database, where images are pre-processed and descriptors are determined using convolutional neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If direct supervised classification is used from rock images, then automation is achieved, but classification precision deteriorates due to errors from not considering naturalistic features
Solution Approach 1:
The patent segments the rock image analysis into multiple independent descriptor analyses (color, texture, shape, etc.) rather than direct classification. Each descriptor is analyzed separately using machine learning, and the results are combined to form the final classification. This segmentation allows the system to capture naturalistic features systematically while maintaining automation.
Solution Approach 2:
The patent introduces intermediate descriptors as mediators between the raw rock image and the final classification. These descriptors (color, texture, shape, etc.) serve as intermediate representations that capture naturalistic features, allowing the system to achieve both automation and precision by analyzing features indirectly rather than directly classifying from the image.
2Measurement precision
If multiple descriptors are analyzed using machine learning, then classification precision is improved, but device complexity increases
Solution Approach 1:
The patent divides the complex classification task into multiple independent descriptor analyses. Each descriptor (color, texture, shape) is processed separately using dedicated machine learning models, which can be trained and optimized independently. This segmentation reduces the overall complexity by breaking down the monolithic classification problem into manageable sub-problems.
Solution Approach 2:
The patent employs universal machine learning techniques (such as convolutional neural networks) that can be applied to multiple different descriptors. The same underlying ML framework handles color analysis, texture analysis, and shape analysis, providing a unified approach that reduces complexity compared to using separate specialized systems for each descriptor.
3Measurement precision
If geologist expertise is used for rock classification, then classification accuracy is improved, but productivity decreases due to manual analysis requirements
Solution Approach 1:
The patent implements a self-service system where machine learning algorithms automatically perform the classification task that previously required geologist expertise. The system analyzes rock images, extracts descriptors, and generates classifications without human intervention, thereby achieving both high accuracy (matching expert level) and high productivity (automated processing of multiple images simultaneously).
Solution Approach 2:
The patent replaces the mechanical system of manual geologist analysis with an automated computational system. Instead of human experts visually examining and classifying rocks, machine learning models process images and descriptors algorithmically, maintaining accuracy while dramatically increasing productivity through automated batch processing.
Data Source
AI summary
The present invention relates to a rock classification method wherein at least one image (IMA) of the rock to be classified is acquired, and wherein a decision tree (ARB) classifying the rocks according to several descriptors is used, as well as a machine learning method (APP) from a rock image database (BIR). Machine learning is applied for each descriptor considered.