DOI number:10.1007/s12583-021-1504-1
Journal:Journal of Earth Science
Key Words:hyperspectral, ion-adsorbed ore, BP neural network, quantitative inversion, REEs
Abstract:Rare earth minerals are important strategic resources to economic development all over the world. In this study, multiple linear regression and back propagation (BP) neural network methods are used to invert the contents of ion adsorbed rare earth elements (REEs) and exploring the feasibility of quantitative inversion of REEs through measured hyperspectral data in Liutang rare earth mines, South China. The result shows that the spectral curve of the rare earth ore samples has obvious absorption characteristics around 390 nm, 930 nm, 1400 nm, 1900 nm and 2200 nm, and continuum removal and the 1st derivative treatment can highlight the absorption characteristics. The modeling accuracies of BP neural network are higher than that of multiple linear regression model. The BP neural network model of the 1st derivative data in 400-1000nm bands has the best inversion result of the total content of REEs, R2 reaches 0.98, the ratio of the performance to deviation (RPD) is larger than 3.0. The quantitative inversion model of each REE (except for Ce) has high precision, R2 is greater than 0.90 and RPD is greater than 3.0. The results indicate that quantitative inversion of REEs using measured spectra not only has great potential and feasibility in the exploration of rare earth minerals, but also provides a rapid test method for the content of ion-adsorbed rare earth elements.
First Author:Gong Cheng
Indexed by:Journal paper
Document Code:001050332100010
Discipline:Engineering
First-Level Discipline:Geological Resources and Geological Engineering
Document Type:J
Volume:34
Issue:4
Page Number:1068-1082
ISSN No.:1674-487X
Translation or Not:no
Included Journals:SCI