AI Image Analysis for Ore Texture, Grain Size and Mineral Mapping

Key Takeaways

• Ore texture analysis, grain size analysis and mineral mapping run on the same automated-mineralogy and hyperspectral imaging stack: TESCAN TIMA, Zeiss Mineralogic, and Corescan for core-scale hyperspectral work.
• These platforms increasingly use CNN-based classifiers rather than fixed spectral lookup tables to resolve grain boundaries and mixed-pixel mineralogy.
• The output feeds two downstream uses directly: comminution and liberation modeling (how finely you need to grind to free the value minerals) and geometallurgical domaining.
• This is a mature, commercially deployed capability, not a research prototype.
• Malvern Panalytical’s Morphologi line is a relevant alternative for particle-scale grain-size distribution work outside the SEM-based platforms.

TL;DR

Ore texture analysis no longer depends on manual petrography. Run polished samples through an automated SEM mineralogy platform (TIMA or Mineralogic), or scan drill core with a hyperspectral scanner (Corescan). You get quantitative texture, grain-size and mineral-map data, classified by built-in machine learning.

What Is Ore Texture Analysis?

Ore texture describes how minerals sit together in the rock: how big the grains are, which minerals touch which, and how finely they are intergrown. Two ores with the same grade can behave very differently in the plant because of texture. Coarse, loosely locked grains break free easily. Fine, tightly intergrown grains need much finer grinding before flotation can recover them.

That makes ore texture analysis a planning tool, not just a description. It tells you how hard the plant will have to work on each part of the deposit. Grain size analysis and mineral mapping are two more views of the same image data, which is why one instrument stack produces all three.

How Do I Analyze Ore Texture, Grain Size and Mineral Mapping With AI?

The choice of tool depends on the scale you need: a polished sample, a whole drill core, or crushed particles.

Automated SEM mineralogy for polished samples

TESCAN TIMA and Zeiss Mineralogic are the primary automated-SEM platforms. They raster an electron beam across a polished mount, collect an EDS spectrum at each point and classify the mineral phase. The data that produces a mineral map also produces grain-boundary segmentation for grain size. It also captures textural relationships such as intergrowths, alteration halos and mineral associations. For a closer look at the three main platforms, see automated mineralogy with TIMA, MLA and QEMSCAN.

Where the AI comes in

What makes this “AI” rather than just automated imaging is the classification step. Modern platforms use trained neural-network classifiers to resolve ambiguous or mixed-pixel spectra. A typical case is a grain boundary where the beam catches two minerals at once. Neural networks handle that more reliably than the fixed spectral lookup tables older systems used. Mineralogic explicitly markets deep-learning-based classification. It matters most in fine-grained or texturally complex ore, where manual point counting would miss detail or take far too long.

Hyperspectral core scanning for ore texture analysis at scale

For core-scale work, Corescan’s hyperspectral imaging scans whole drill core directly. It extracts mineralogical and textural information from reflectance spectra without cutting and mounting samples. That makes it useful for fast, non-destructive screening of a whole hole before choosing which intervals deserve detailed SEM work. The same approach underpins core scanning more broadly.

Malvern Panalytical’s Morphologi instruments complement both. They measure particle-scale grain size distribution, which is useful after crushing or grinding, when you are characterising the product rather than the ore in place.

How Ore Texture Analysis Feeds the Plant

The practical payoff comes in two places. First, grain size and liberation data tell you how finely to grind the ore to expose value minerals for flotation. That directly shapes grinding circuit optimization. Second, texture and mineral-map data feed geometallurgical domaining, which groups the deposit into zones that process in similar ways. This is a mature capability across the board. You are choosing between well-established commercial instruments, not evaluating unproven technology.

FAQs

Why does ore texture matter for processing?

Ore texture controls how easily valuable minerals separate from waste when the rock is crushed and ground. Coarse, simple textures liberate at a coarser grind, which saves energy. Fine, complex intergrowths need finer grinding and can still lose metal to tailings. Knowing the texture up front helps size the grinding circuit and predict recovery.

What is the difference between TIMA and Mineralogic?

Both are automated SEM mineralogy platforms that classify minerals from EDS spectra and produce mineral maps, grain size and texture data. TIMA comes from TESCAN and Mineralogic from Zeiss. The practical differences lie in hardware, software workflow and classification options, so most labs choose based on existing equipment and support.

Can hyperspectral scanning replace SEM mineralogy?

Not entirely. Hyperspectral core scanning is fast and non-destructive and covers whole drill holes, but it works at a coarser scale and detects a different set of minerals. SEM mineralogy gives grain-level detail on polished samples. Most programs use hyperspectral scanning to screen and SEM mineralogy on the intervals that matter most.

How accurate is AI mineral classification?

Neural-network classifiers resolve mixed or ambiguous spectra more reliably than fixed lookup tables, especially at grain boundaries and in fine-grained ore. Accuracy still depends on good training data and calibration for your specific ore, so check results against manual petrography on a subset of samples.

Try It With Geocluster

The real work is connecting texture, grain-size and mineral-map results back to your block model and geometallurgical domains so they inform mine planning decisions. That cross-dataset research is what Geocluster is built to support. If you are building geometallurgical domains from automated mineralogy data, it can help connect the pieces.