Segmentation of mine overburden dump particles from images using Mask R CNN

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence in Mining
• Image Analysis for Particle Characterization

Background:

• Mine overburden dump stability is critical for mining operations.
• Particle size distribution significantly influences dump slope shear strength.
• Automated identification of dump particles from images is challenging due to variations in visual characteristics.

Purpose of the Study:

• To present a novel Artificial Intelligence (AI) approach for identifying particles in mine dump images.
• To determine particle size distribution of mine dumps using AI-driven image analysis.
• To automate the process of particle boundary detection and coordinate calculation.

Main Methods:

• Implementation of Mask R-CNN with ResNet50 and Feature Pyramid Network (FPN) for instance segmentation.
• Pixel-level segmentation of individual particles from dump images.
• Training the model on a dataset of 31,505 particles.

Main Results:

• The AI model achieved a training accuracy of 97.2%.
• The model accurately delineated particles and generated masks with a mean percentage error of 0.39% compared to ground truth.
• A standard deviation of 0.25% was observed in area prediction, indicating high precision.

Conclusions:

• The developed AI model effectively segments mine dump particles at a pixel level.
• This method provides a significant reduction in time and effort for particle size distribution analysis in rock mechanics.
• The approach offers a promising solution for automated particle identification and characterization in mining.