Ore identification & pre-sorting
Build material-specific models from spectral differences and use RGB localization to support rejection, grading and routing.
MINING / SPECTRAL SORTING
Read the material.
Recover its value.
Spectral identification, sorting integration and environmental monitoring for mines and material recovery operations.
01 / OUR EXPERIENCE
Iron ore classification research
Zhongke Yaoguang has carried out classification work on the two iron-ore sample categories referred to as “精铁矿” and “粗铁矿”. This work provides a starting point for application-specific spectral identification.
The next stage connects representative samples, laboratory assays and independent-batch validation to conveyor-based testing.
Research focusIron ore sample classificationDevelopment pathSpectral identification → Sorting integration
02 / APPLICATIONS
One sensing platform, connected tasks
Waste classification & recovery
Identify components in demolition waste, mine waste rock and mixed streams, selecting visible, NIR or SWIR sensing for the target materials.
Mine water & site monitoring
Combine fixed water analyzers, portable spectroscopy, UAV payloads and AI cameras for outfall monitoring, water changes and site inspection.
03 / SYSTEM INTEGRATION
A practical route to sorting
Controlled acquisition
Metered feed, stable illumination and dark/white references support spectral and image capture.
Features & decisions
Select informative bands and combine texture and shape to report class, position and confidence.
Actuation & verification
Link encoders, PLCs and actuators; measure recovery, purity and misclassification.
Yaoguang supplies sensing, recognition and integration interfaces. Conveyor mechanics and sorting actuators are configured with equipment partners. Wavelength coverage, particle size, moisture and belt speed determine the test plan.
04 / PUBLISHED RESEARCH
Published experiments & measured data
The following results are attributed to their authors and experimental settings; they inform sensor selection and test design.
F1 · Mine-face classification
Martite, shale and sky background; the combined transfer-learning and augmentation method uses 430–860 nm. Images were acquired at 11:30 and 13:30. This is mine-face mapping, not conveyor recovery.
Read study ↗Overall accuracy · PCA–ELM
The study classifies plastic, foam, wood, brick and concrete with a 900–1700 nm sensor. The result belongs to the reported experimental dataset and classifier.
Read study ↗mIoU · Waste segmentation
SpectralWaste Table IV: SegFormer-B0 with RGB versus CMX-B0 with RGB–HSI hybrid fusion. Both input and architecture change. The dataset contains 852 labeled images and 2,059 instances across six classes.
Read study ↗05 / SPECTRAL MEASUREMENTS
Reflectance measurements from published samples
mortar-01mortar-02mortar-03mortar-04
Four mortar samples / 400–1000 nm
The chart redraws seven measured reflectance features per sample from Appendix C, Table 1 of Vítek et al. (2025). Lines connect tabulated points without interpolation of additional measurements.
The same study identifies 500 and 800 nm as useful additions to RGB features for its demolition-waste classification task. This offers a reference for selecting task-specific metasurface channels.
Source: Appendix C, Table 1 ↗Download tabulated values (CSV)
View tabulated values
| nm | mortar-01 | mortar-02 | mortar-03 | mortar-04 |
|---|---|---|---|---|
| 400 | 0.4100 | 0.4011 | 0.3763 | 0.4198 |
| 500 | 0.4642 | 0.4623 | 0.4181 | 0.4564 |
| 600 | 0.6407 | 0.6266 | 0.5234 | 0.5143 |
| 700 | 0.6757 | 0.6658 | 0.5466 | 0.5325 |
| 800 | 0.6905 | 0.6871 | 0.5547 | 0.5392 |
| 900 | 0.6842 | 0.6870 | 0.5496 | 0.5438 |
| 1000 | 0.6884 | 0.6950 | 0.5494 | 0.5528 |
06 / ENGINEERING
From a sample to a site trial
Sample screening
Collect representative batches, assay labels and dry/wet surface conditions.
Bench validation
Measure spectral separation, select bands and evaluate held-out batches.
Conveyor trial
Integrate synchronization and actuation; evaluate material recovery, purity and throughput.
Discuss your material and sorting target
Send material types, particle size, moisture and throughput targets.
BD@zhongkeyaoguang.com ↗- Windrim et al. · 2023. Unsupervised ore/waste classification on open-cut mine faces using close-range hyperspectral data.
- Xiao et al. · 2019. Development of online classification system for construction waste based on industrial camera and hyperspectral camera.
- Casao et al. · 2024. SpectralWaste Dataset: Multimodal Data for Waste Sorting Automation.
- Vítek, Zbíral & Nežerka · 2025. Identification of critical wavelengths for classifying construction and demolition waste materials using hyperspectral imaging.