🎓 Lesson 4
D3
Hyperspectral Signatures of Key Ore Minerals
Hyperspectral signatures are like unique 'light fingerprints' that different ore minerals leave when scanned with special cameras that see hundreds of narrow color bands.
🎯 Learning Objectives
- ✓ Analyze hyperspectral reflectance curves to identify diagnostic absorption features of hematite, chalcopyrite, and sphalerite
- ✓ Apply continuum-removed spectral preprocessing to isolate and compare absorption depths of key minerals
- ✓ Explain how grain size, surface roughness, and Fe³⁺/Fe²⁺ ratio affect the position and shape of the 860–920 nm hematite absorption feature
- ✓ Design a field-acquired hyperspectral survey protocol for grade control that meets ISO 18436-3 spectral fidelity requirements
📖 Why This Matters
In AI-powered orebody delineation, knowing *what* mineral is present—and *how much*—is useless without knowing *where it is* at sub-meter resolution in real time. Hyperspectral imaging bridges the gap between lab-grade XRD/ICP assays and operational decision speed: a single drone flight over a blast face can map goethite vs. hematite distribution, flag low-grade dilution zones before loading, and feed grade-control models with pixel-level mineral abundance—not just bulk geochemistry. Mistaking a clay-rich waste zone for oxidized copper ore costs millions in processing; this lesson equips you to detect that error before the shovel moves.
📘 Core Principles
Minerals absorb light at specific wavelengths due to quantum-mechanical interactions: electronic transitions (e.g., Fe²⁺→Fe³⁺ in hematite near 860 nm), OH-stretch overtones (e.g., kaolinite at 1400, 1900, 2200 nm), and metal–O–H bending modes (e.g., chlorite at 2340 nm). Hyperspectral sensors capture 100–300+ contiguous bands—unlike multispectral sensors (3–10 broad bands)—enabling detection of subtle, narrow absorption features. Critical concepts include spectral resolution (bandwidth), sampling interval (band center spacing), signal-to-noise ratio (SNR > 300:1 required for mineral ID), and atmospheric correction (especially for SWIR water vapor bands at 1360–1410 nm and 1890–1940 nm). Mineral identification relies on matching observed absorption minima (center wavelength, depth, width, asymmetry) to reference libraries such as the USGS Spectral Library v7 or ECOSTRESS Mineral Database.
📐 Continuum Removal
Continuum removal normalizes spectra to isolate absorption features by removing background slope and curvature—essential for comparing feature depth across samples. It expresses absorption strength as a unitless ratio, enabling quantitative mineral abundance estimation via partial least squares regression (PLSR) or spectral angle mapper (SAM) classification.
Continuum Removal
CR(λ_i) = 1 − \frac{R(λ_i)}{R_{cont}(λ_i)}Normalizes spectral reflectance to quantify absorption feature depth independent of overall brightness.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| CR(λ_i) | Continuum-removed reflectance | unitless | Depth of absorption feature at wavelength λ_i |
| R(λ_i) | Measured reflectance | unitless | Raw sensor reflectance at wavelength λ_i |
| R_{cont}(λ_i) | Continuum reflectance | unitless | Interpolated baseline reflectance at λ_i, derived from convex hull endpoints |
Typical Ranges:
Strong hematite absorption: 0.30 – 0.65
Weak sphalerite Zn–S stretch: 0.08 – 0.22
💡 Worked Example
Problem: A hyperspectral scan of a drill core sample shows reflectance values at 850 nm = 0.32, 875 nm = 0.21 (absorption trough), and 900 nm = 0.34. Calculate the continuum-removed absorption depth at 875 nm.
1.
Step 1: Fit a straight-line continuum between endmembers: R_cont(λ) = R(850) + [(R(900) − R(850)) / (900 − 850)] × (λ − 850)
2.
Step 2: Compute continuum value at 875 nm: R_cont(875) = 0.32 + [(0.34 − 0.32)/50] × 25 = 0.32 + 0.01 = 0.33
3.
Step 3: Apply continuum removal: CR(875) = 1 − [R(875) / R_cont(875)] = 1 − (0.21 / 0.33) = 1 − 0.636 = 0.364
Answer:
The absorption depth is 0.364 (unitless), indicating moderate hematite abundance—consistent with a 15–25 wt% hematite zone per USGS spectral calibration curves.
🏗️ Real-World Application
At Rio Tinto’s Koodaideri iron ore mine (Pilbara, WA), hyperspectral pushbroom sensors (Headwall Nano-Hyperspec® SWIR, 274 bands, 5.4 nm FWHM) were mounted on autonomous haul trucks to scan freshly blasted faces in real time. By targeting the 860 nm (Fe³⁺ charge transfer) and 660 nm (crystal field) hematite absorptions—and suppressing interference from goethite using the 1400 nm H₂O/OH band ratio—the system achieved 92% accuracy in distinguishing high-grade (>62% Fe) hematite domains from interbedded jasperoid (≤55% Fe). This reduced re-handling by 18% and increased mill feed grade consistency (σFe < 0.8%)—directly feeding Rio’s AI grade-control engine 'MineSense'.
🔧 Interactive Calculator
🔧 Open AI-Powered Orebody Delineation & Grade Control Calculator📋 Case Connection
📋 Gold Mine Real-Time Grade Control at Development Drift Face
Manual chip sampling caused 24–48 hr delay in stope boundary decisions, leading to 14% dilution
📋 Limestone Mine Digital Twin for Karst-Related Grade Uncertainty
Solution cavities caused unpredictable grade drops (CaCO₃ purity <85%) in otherwise uniform deposits, resulting in 11% p...