Test Case | Hyperspectral ATH1010/ATH8100 Series Ore Testing

Traditional ore sorting mainly relies on manual experience, colour recognition, and conventional physical testing, which suffers from low efficiency, strong subjectivity, and limited detection accuracy. Different minerals vary in chemical composition, crystal structure, and hydration state, producing distinct absorption, reflection, and spectral response characteristics at specific wavelengths. Hyperspectral imaging technology can acquire continuous, detailed spectral information from ores, enabling simultaneous detection of “spatial position + spectral features” and providing a technical foundation for rapid, non‑destructive, and intelligent ore sorting.
 

This solution addresses the needs of ore category identification, grade classification, and impurity removal. By acquiring spectral data from ore samples via hyperspectral imaging, combined with spectral preprocessing, feature extraction, machine learning, and intelligent classification algorithms, an ore spectral identification model is established to achieve precise discrimination among different ores, different grades, and waste rock.

01  Principles of Ore Identification
 

Elements such as iron, copper, magnesium, aluminium, and silicon in ores, as well as different mineral crystal structures, produce characteristic absorption or reflection at specific wavelengths of electromagnetic radiation. For example, iron‑bearing minerals typically exhibit obvious visible‑near‑infrared spectral features, while minerals containing hydroxyl and carbonate groups also show typical absorption features in the short‑wave infrared region.
 

Granite spectral curves
 

The system uses a hyperspectral imager to continuously scan ores on a conveyor belt or sample stage. Each piece of ore generates a corresponding spectral curve and hyperspectral data cube. By comparing spectral curves of different ores, distinctive feature bands are extracted, and classification models are built using algorithms such as PCA, SPA, Random Forest, SVM, and PLS to determine ore category, quality, and anomalous impurities.

Compared with plain RGB images, hyperspectral technology utilises information from hundreds of contiguous bands. Even when ore surfaces have similar colours, identification can be achieved based on spectral differences arising from internal material composition.

02  System Composition and Sorting
 

The entire intelligent ore sorting system mainly consists of a feeding and conveying system, a hyperspectral imaging system, a light source system, data acquisition and analysis software, intelligent recognition algorithms, and sorting actuators.
 

Hyperspectral camera installed on sorting line

Ore first enters the detection area uniformly via the conveyor belt. The hyperspectral imager rapidly scans the ore while simultaneously collecting corresponding visible, near‑infrared, or short‑wave infrared spectral information. After the system completes preprocessing such as dark current correction, white reference correction, denoising, and spectral normalisation, it extracts the target region of the ore and calculates average spectra, characteristic spectra, and related spectral indicators.
 

Recorded spectral curves

Subsequently, the trained recognition model automatically classifies the ore. Based on the recognition results, spatial positions and sorting instructions for different categories are generated. Finally, air jets, robotic arms, baffles, or other actuators send different types of ore into corresponding bins, achieving automated sorting.
 

Classification results

Basic Workflow:

Ore feeding → Conveyor transport → Hyperspectral scanning → Data preprocessing → Spectral feature extraction → Classification and recognition → Target positioning → Sorting execution → Category collection.

03  Advantages and Application Value

Hyperspectral ore sorting offers advantages such as speed, non‑destructiveness, non‑contact operation, high information dimensionality, and a high degree of intelligence. The system simultaneously acquires spatial information and continuous spectral information from ores, upgrading from traditional “colour‑based” sorting to “spectrum‑based” sorting, and enhancing recognition capability for similar minerals and complex ore environments.

By establishing standard spectral databases for different ores and different grade samples, it is possible to further achieve ore category identification, grade classification, waste rock removal, and mixed mineral screening, with models continuously optimised according to actual production needs.

This solution can be applied to intelligent detection and sorting of iron ore, copper ore, bauxite, lithium ore, phosphate ore, coal gangue, and other mineral resources. It provides technical support for mining enterprises to improve resource utilisation, reduce labour costs, and decrease waste rock entering subsequent processing stages – driving ore sorting toward digitalisation, automation, and intelligent development.

For more information, please contact:

Email: optoskyphotonics@gmail.com

Web: www.optosky.net