关键词: 高光谱　, 稀土矿　, 重金属竞争性　, 特征筛选　, 多目标优化　, Lotka-Volterra模型　, XGBoost

Abstract: Hyperspectral technology has the potential to retrieve heavy metal elements in soils of rare earth mining areas. However,complex competitive adsorption between rare earth elements (La,Ce) and heavy metals (Cu,Pb) in soil components leads to overlapping and masking of spectral response features,limiting retrieval accuracy.To address this issue,a typical rare earth mining area in southern Jiangxi Province was selected as the study area,and 153 soil samples were collected.Based on an in-depth investigation of the spectral competition relationships between rare earth elements and heavy metals,the Lotka-Volterra model was employed to quantitatively characterize the competition intensity among elements.Accordingly,a multi-objective hy perspectral band selection method incorporating the mechanism of spectral competition among soil heavy metals was proposed to effectively select characteristic spectral bands and construct band ratios.Additionally,terrain factors (e.g.,elevation,slope, curvature) and mining factors (distance from sampling points to sedimentation ponds) were introduced to establish a hyper spectral quantitative retrieval model using the XGBoost algorithm.The results show a significant competitive adsorption effect between Cu-La-Ce and Pb (p<0.01).The maximum correlation between the constructed characteristic indices and soil heavy metal contents significantly improved (Cu:from 0.32 to 0.70;Pb:from 0.33 to 0.49).After incorporating terrain and mining factors,the model accuracy further increased,with the full-variable model achieving validation set R2 values of 0.89 for Cu and 0.78 for Pb,representing improvements of 14.3% and 6.5%,respectively,compared to the sub-model using only band ratio features.This study provides a new approach for hyperspectral monitoring of heavy metal pollution in rare earth mining areas.

Key words: hyperspectrum,rare earth mine,heavy metal competition,feature selection,multi-objective optimization,Lot ka-Volterra model,XGBoost