ABSTRACT

Hyperspectral remote sensing provides apromising solution for estimation of heavy metal concentration in soil. However, few studies have focused on the detection of soil heavy metal concentration by the frequency domain analysis. In this letter, the Hilbert–Huang transform (HHT) is introduced to fully explore the hidden information in the spectrum. Based on experimentally acquired spectra of soil contaminated by lead (Pb) and chemical data, HHT was employed to obtain the Hilbert energy spectra (HES) and intrinsic model function (IMF) component of spectra. Then, characteristic spectral bands of Pb could be fully mined through these components, and random forest was utilized to retrieve Pb concentration. The following conclusions are drawn: (1) the components of HHT has good correlation with Pb concentration in the 340–1400 nm, and they can better highlight response of Pb concentration in the 1440–2450 nm; (2) characteristic bands extracted by the IMF components and HES are quite effective as input variables, and its correlation coefficient (r) and root mean square error (RMSE) for random forest is 0.9676 and 0.0741, respectively. Compared with the variables of original spectral reflectance and spectral domain transformation, the proposed spectral HHT variables achieve the highest estimation accuracy.

摘要

高光谱遥感为估计土壤中的重金属浓度提供了有希望的解决方案。然而，很少有研究关注通过频域分析检测土壤重金属浓度。在这封信中，引入了希尔伯特-黄变换（HHT）以充分探索光谱中的隐藏信息。基于实验获得的铅(Pb)污染土壤光谱和化学数据，利用HHT获得了光谱的希尔伯特能谱(HES)和本征模型函数(IMF)分量。然后，通过这些成分可以充分挖掘出 Pb 的特征光谱带，并利用随机森林来检索 Pb 浓度。得出以下结论：（1）HHT的成分与340-1400 nm范围内的Pb浓度具有良好的相关性，它们可以更好地突出 1440-2450 nm 范围内 Pb 浓度的响应；(2) IMF分量和HES提取的特征带作为输入变量相当有效，其相关系数为(r ) 和随机森林的均方根误差 (RMSE) 分别为 0.9676 和 0.0741。与原始光谱反射率和光谱域变换的变量相比，所提出的光谱HHT变量达到了最高的估计精度。