铅锌矿集区土壤重金属含量高光谱反演方法研究  

Research on Hyperspectral Retrieval Method for Heavy Metals Content of Soil in Lead-Zinc Mining Areas

作　　者：李会玲 刘严松[1,2] 沈杜衡 薛凯伦 聂诗音 唐方强 LI Huiling;LIU Yansong;SHEN Duheng;XUE Kailun;NIE Shiyin;TANG Fangqiang(Key Laboratory of Earth Exploration and Information Techniques(Chengdu University of Technology),Ministry of Education,Chengdu University of Technology,Chengdu,Sichuan 610059,China;Sichuan Sanhe Space Technology Co.,Ltd.,Chengdu,Sichuan 610094,China)

机构地区：[1]成都理工大学地球勘探与信息技术教育部重点实验室,四川成都610059 [2]四川三合空间科技有限公司,四川成都610094

出　　处：《矿业研究与开发》2025年第2期204-212,共9页Mining Research and Development

基　　金：国家自然科学基金项目(41971226);中国地质调查局地调项目(DD20221697);四川省自然资源厅基金项目(KJ2016-16);四川省教育厅基金项目(18ZB0065);甘肃省教育厅高校教师创新基金项目(2023A-253)。

摘　　要：铅锌矿区存在严重的重金属复合污染问题,可利用高光谱遥感技术,提取土壤光谱信息中重金属吸收特征波谱,实现矿区土壤重金属含量反演。以西藏斯弄多铅锌矿集区土壤为研究对象,基于高光谱开展了土壤重金属元素(Cd、Pb、As、Hg)含量的反演研究。对土壤样品进行高光谱数据采集,并进行一阶微分(FD)、二阶微分(SD)、倒数对数(AT)、倒数对数一阶微分(AFD)、倒数对数二阶微分(ASD)多种光谱变换,筛选出与重金属实测含量相关性较强的特征波段,最后建立多元逐步回归(SMLR)、支持向量机(SVM)、人工神经网络(ANN)、随机森林(RF)4种反演模型,选取决定系数(R2)和均方根误差(RMSE)评价模型精度,得出各重金属的最优反演模型。研究结果表明:(1)不同光谱变换方式的数据降维效果与特征波段筛选区间不同,5种变换中SD变换效果最好,能够有效区分出特征波段,其次为ASD、AFD变换;(2)RF模型反演效果最好,适用性和反演精度优于SMLR、ANN和SVM;(3)As的最佳反演模型为AT-RF模型,Cd的最佳反演模型为SD-RF模型,Pb的最佳反演模型为ASD-RF模型,Hg的最佳反演模型为ASD-SMLR模型;(4)土壤的Cd、Pb、As、Hg含量模型评价均值超出西藏土壤背景值的124.2,89.8,0.70,1.24倍,说明该矿集区的土壤以Cd和Pb为主要污染因子,同时伴有As和Hg的复合污染。There is a serious problem of heavy metal compound pollution in lead-zinc mining areas.Hyperspectral remote sensing technology can be used to extract heavy metals absorption characteristic spectra from soil spectral information and achieve inversion of heavy metal contents in the soil of mining areas.Based on hyperspectral data,the heavy metals contents in soil,such as Cd,Pb,As and Hg,were retrieved from the soil of the Silongduo lead-zinc mining area in Xizang.The hyperspectral data from soil samples was collected and various spectral transformations were performed,including first differential(FD),second differential(SD),logarithm of the reciprocal(AT),first differential of logarithm of the reciprocal(AFD),and second differential of logarithm of the reciprocal(ASD).The characteristic bands that are strongly correlated with the measured content of heavy metal were selected and finally four inversion models were established,including stepwise multiple linear regression(SMLR),support vector machine(SVM),artificial neural network(ANN),and random forest(RF).The coefficient of determination(R~2)and root mean square error(RMSE)were selected to evaluate the accuracy of the models and the optimal inversion model for each heavy metal was obtained.The research results show that the dimensionality reduction effect and feature band selection interval of different spectral transformation methods are different.Among the five transformations,SD transformation has the best effect and can effectively distinguish feature bands,followed by ASD and AFD transformations.The RF model has the best inversion effect,with a better applicability and inversion accuracy than SMLR,ANN,and SVM.The best inversion model for As is the AT-RF model,the best inversion model for Cd is the SD-RF model,the best inversion model for Pb is the ASD-RF model,and the best inversion model for Hg is the ASD-SMLR model.The average assessment values of Cd,Pb,As,Hg content models of soil exceeded the background value of Xizang soil by 124.2,89.8,0.70,1.24 times,in

关 键 词：高光谱反演 重金属元素 特征波段 含量预测 铅锌矿 

分 类 号：TD163.2[矿业工程—矿山地质测量] X833[环境科学与工程—环境工程]