GemstoneClassificationAI
Scaling industrial R&D with spatial intelligence.
A customized AI research pipeline for high-volume gemstone identification, designed to streamline quality control and preliminary sorting for gemstone mining operations.
Project Information
Client
Gems & Minerals Mining Corp
Year
2025
Role
AI R&D Partner
Scope
- Computer Vision
- Deep Learning
- R&D
Overview
Developed a systematic AI framework to automate 68 distinct gemstone classifications, enabling the client to scale their sorting operations and maintain consistent quality standards from the mining site.
The Challenge
The client's sorting department struggled with the slow, manual classification of raw stones. Dependency on human experts limited their ability to process massive extraction volumes and led to quality control bottlenecks.
The Solution
Implemented a pipeline with smart pre-processing (Otsu thresholding) and feature extraction (color histograms, texture analysis). Trained a Random Forest model that outperformed deep learning alternatives for this specific use case.
Technology Stack
Language
- Python
Vision
- OpenCV
- ResNet
ML
- Random Forest
- Scikit-learn
Key Impact
Processing time reduced from 175 minutes (human) to under 1 second (AI). Achieved 69.4% accuracy, surpassing expert benchmarks. Scalable to 68 gemstone types.
Methodology Framework
RGB Color Space Analysis
HSV Feature Visualisation
Classification Accuracy Boxplot
Best System Confusion Matrix
Expert Gemmologist Confusion Matrix
"This system transformed our field operations. It not only saves 99% of screening time for raw stones but also provides the objective classification data we need for industrial scaling."
