基于高光谱的土壤重金属含量预测模型建立与评价

Establishment and Evaluation of Prediction Model for Heavy Metal Content Based on Hyperspectral Data

研究不同重金属元素的统计特性以及与光谱不同变换形式之间的相关性,通过逐步回归算法和最佳适应值函数F为指标选取最佳波段,运用偏最小二乘回归方法构建了不同变换形式的光谱反射率与重金属含量的反演模型。结果表明:研究区内Zn、Cu和Ni主要受自然因素影响,Cr受外界因素影响程度较大且存在一定的积累现象。不同光谱变换方式的建模精度和预测能力大小有以下关系,光谱对数微分>光谱一阶微分>光谱倒数微分>光谱连续统去除>光谱倒数对数>原始光谱。采用光谱对数一阶微分建模可以作为反演研究区土壤重金属含量的最佳模型,从而为研究区土壤重金属含量快速监测和大尺度的土壤重金属污染评价提供技术支撑。

This paper mainly studies the statistical properties of different heavy metals and the correlations between spectrum reflectance under different transformations and the content of heavy metals, and models between the spectrum reflectance and the content of heavy metals by the stepwise regression algorithm and the best fitness function F as the index to select optimal bands. The results show that Zn, Cu and Ni are influenced by natural factors, that Cr is influenced by human activities and that there is a phenomenon of accumulation to a certain extent. The following re- lationship exists in the modeling accuracy and predictive power of spectrum reflectance under different transformations： logarithmic differential 〉 first-order differential 〉 inverse differential 〉 continuum removal 〉 inverse logarithms 〉 original spectra. First-order differential spectral logs can be modeled to invert the content of heavy metals in the soil, which may offer technical support for fast detection of the content of heavy metals and pollution assessment of soil that contains large scales of heavy metals.