In recent years, the technology of visible and near infrared spectral inversion of heavy metals in soil of a mining area has been attracting more and more attention. However, the contents of heavy metals in the soil are often so trivial that their spectral characteristics are very fragile and hence the requirements of their inversions and for the models should be much higher. In a study on inversions of heavy metals in the soil of reclaimed mining areas, the technology of extreme learning machine ( ELM) was introduced to inversion modeling and compared with the traditional partial least squares re⁃gression( PLS) and the support vector machine ( SVM) methods. After pretreatment and correlation analysis of spectral da⁃ta, the three models were used to inverse the data of 30 soil samples, and 10 of them were chosen for model validation. Results show that the model of ELM was higher than the models of SVM and PLS inaccuracy of the prediction of Zinc ( Zn) , Copper ( Cu) , Cadmium ( Cd) and Chromium ( Cr) and more or less the same in prediction capacity for Plumbum ( Pb) and Arsenic ( As) with SVM.%近年来采用可见光近红外光谱反演矿区土壤重金属受到重视，但土壤中重金属含量微小，光谱特性非常脆弱，对反演模型提出了较高要求。针对复垦矿区的土壤重金属反演研究，引入超限学习机（ extreme learning machine，ELM）方法进行反演建模，与传统的偏最小二乘（ partial least squares regression， PLS）方法和支持向量机（ support vector machine， SVM）方法进行分析比较。通过对光谱数据进行预处理和相关性分析后，对30个土壤样本数据运用3种模型进行反演，并对其中10个预测样本进行模型检验。结果表明，ELM对于重金属Zn、Cr、Cd和Cu的预测精度要高于SVM和PLS，对重金属As和Pb的预测能力与SVM基本相当。
展开▼
