A Photomicrographic Dataset of Rocks for the Accurate Classification of Minerals

Area of Science:

• Geological Sciences
• Mineralogy
• Computer Vision

Background:

• Automated mineral identification in petrography is hindered by insufficient datasets covering full optical characteristics.
• Existing datasets often lack comprehensive data on orientation-dependent optical properties.

Purpose of the Study:

• To introduce the Menoufia University Machine Learning Dataset for Minerals Classification 2025 (MUMDMC2025).
• To provide a comprehensive dataset for machine learning-based mineral classification in petrography.
• To address the challenge of limited optical data in mineral identification.

Main Methods:

• Collected 14,400 high-resolution photomicrographs of five mineral classes (Biotite, Hornblende, Plagioclase, Potassium-Feldspar, Quartz).
• Imaged each specimen across 72 rotational positions (360° coverage) under Plane Polarized Light (PPL) and Cross Polarized Light (XPL).
• Documented complete anisotropic optical properties including pleochroism, birefringence, and extinction patterns.

Main Results:

• The MUMDMC2025 dataset is balanced, with 2,880 images per mineral class.
• Dataset validation using K-Nearest Neighbors achieved high classification accuracy.
• The dataset captures orientation-dependent optical phenomena crucial for reliable classification.

Conclusions:

• The MUMDMC2025 dataset significantly enhances capabilities for automated petrographic analysis.
• It supports the development of advanced machine learning models for quantitative mineralogy.
• The dataset is valuable for educational applications in optical mineralogy and geological research.