Hyperspectral inversion of rare earth element concentration based on SPA-PLSR model

Abstract

Quantitative study of the relationship between the hyperspectral characteristics of carbonatite rare earth elements and their chemical concentration is of great significance for detecting carbonatite rare earth resources using remote sensing hyperspectral technology. Due to the high resolution and large number of bands in hyperspectral data, it is crucial to effectively extract characteristic spectral bands with a high correlation with rare earth element concentration for estimating rare earth element concentration based on hyperspectral data.Thirty-three samples of rare earth ore were collected from the Maoniuping rare earth ore district, and indoor hyperspectral measurements were conducted using SVC HR1024I ground-based spectrometer. The Cerium(Ce) element concentration was chemically analyzed by ICP-MS. To improve the accuracy of the spectral inversion model and minimize the interference of stray light, noise, baseline drift, etc., the original spectral data were resampled at intervals of 10 nm first, and then the resampled results were subjected to first-order derivative (FD), Savitzky-Golay smoothing filtering(SG), standard normal variate transformation(SNV), multivariate scattering correction (MSC), and first-order derivative followed by SG filtering(FD_SG) transformations. Based on the successive projection algorithm (SPA), only five to nine selected characteristic bands out of 216 bands ranging from 350 nm to 2500 nm were extracted, reducing the band number by 95.8% to 97.7%, greatly reducing the redundancy of the spectrum. The partial least square regression (PLSR) model constructed based on the characteristic bands selected by SPA and the measured Ce element concentration showed that the determination coefficient(R²) and root mean square error(RMSE) of the modeling set were 0.88 and 363 × 10^–6, respectively, while those of the prediction set were 0.87 and 503 × 10^–6, respectively, indicating good stability and high precision of the model, which can be used as an estimation model for the Ce element concentration in the Maoniuping rare earth ore district.