Comprehensive evaluation of the Segment Anything Model (SAM) for post-blast rock fragmentation analysis: a multi-method comparative study integrating data intelligence at an open-pit gold mine

Authors

DOI:

https://doi.org/10.17159/

Keywords:

rock fragmentation, deep learning segmentation, particle size distribution, mine-to-mill, grizzly feeder, open-pit mining

Abstract

Accurate assessment of post-blast rock fragmentation is critical for optimizing the mining value chain. Conventional methods such as sieve analysis are time-consuming, while commercial image-analysis software requires costly licenses and manual intervention. This study presents a multi-method evaluation of Meta’s Segment Anything Model for automated segmentation and particle size distribution analysis of post-blast rock fragments at an open-pit gold mine in Sumatra, Indonesia. Four complementary investigations were conducted: a laboratory-scale comparison against Split Desktop, a field-scale comparison against Strayos with excavator digging time correlation, a pixel-level evaluation against WipFrag using drone-captured imagery, and a statistical analysis of fragmentation metrics versus grizzly feeder clean-up time. The Segment Anything Model 2.1 achieved a mean intersection-over-union of 0.75 and Dice coefficient of 0.85 on high-quality patches, with particle size distribution errors reduced by 89-99% relative to WipFrag (up to 90 times lower). Linear regression revealed a significant correlation between maximum fragment diameter and grizzly feeder clean-up time (). Analysis across 24 drone images identified over 23,000 fragments, confirming that more than 99% of blasted material passed the 750 mm feeder aperture. Processing time was reduced from 30 minutes to 2–5 minutes per image. These findings validate the Segment Anything Model as a viable open-source data intelligence tool for rock fragmentation analysis, advancing data intelligence and operational risk management within the mine-to-mill optimization framework.

Downloads

Download data is not yet available.

Published

2026-11-06

Issue

Section

Geometallurgy 2026 Themed Edition