Within-Deposit Visible-Near-Infrared Hyperspectral Estimation of Lead Grade in Complex Galena-Bearing Ore: Full-Spectrum Chemometrics, Attribution-Informed Descriptors, and X-Ray Diffraction-Guided Unmixing

کد مقاله : 1109-IMEC15 (R1)

نویسندگان

Nima Jafari *، Mahdi Gharabaghi، Sied Ziaedin Shafaei Tonkaboni، Mahdi Hasanlou دانشگاه تهران

چکیده مقاله

Fast, non-destructive estimation of lead grade can support grade control, feed management, and sensor-based sorting. This study evaluated 43 lead-zinc ore samples from the Changarzeh deposit (0.09-69.27 wt.% Pb) using visible-near-infrared (VNIR) hyperspectral imaging. Three representations were compared under the same leave-one-sample-out validation: the full 992-band spectrum, a predefined set of spectral descriptors, and X-ray diffraction (XRD)-guided spectral unmixing. Full-spectrum partial least squares regression (PLSR) achieved R² = 0.915 and RMSE = 5.60 wt.% Pb. Random forest (RF) applied to the descriptor representation gave the lowest overall point error (R² = 0.938; RMSE = 4.78 wt.% Pb), while global-unmixing gradient boosting reached R² = 0.848. The models showed different grade-dependent error patterns: descriptor RF was especially accurate for low-grade samples, whereas PLSR was more stable in the medium- and high-grade ranges. The results show that VNIR spectra contain useful indirect information for within-deposit Pb-grade screening, but provenance-balanced external validation is still required before transfer to field applications

کلیدواژه ها

VNIR hyperspectral imaging; lead-grade estimation; PLSR; random forest; spectral descriptors; XRD-guided unmixing

وضعیت: پذیرفته شده برای ارائه شفاهی