Hyperspectral indirect inversion of heavy-metal copper in reclaimed soil of iron ore area

Qiang Shen; Ke Xia; Shiwen Zhang; Chenchen Kong; Qingqing Hu; Shaowei Yang

doi:10.1016/j.saa.2019.117191

Hyperspectral indirect inversion of heavy-metal copper in reclaimed soil of iron ore area

Spectrochim Acta A Mol Biomol Spectrosc. 2019 Nov 5:222:117191. doi: 10.1016/j.saa.2019.117191. Epub 2019 Jun 6.

Authors

Qiang Shen¹, Ke Xia¹, Shiwen Zhang², Chenchen Kong³, Qingqing Hu³, Shaowei Yang¹

Affiliations

¹ School of Geodesy and Geomatics, Anhui University of Science and Technology, Huainan 232001, China.
² College of Earth and Environmental Science, Anhui University of Science and Technology, Huainan 232001, China. Electronic address: swzhangaust@163.com.
³ College of Earth and Environmental Science, Anhui University of Science and Technology, Huainan 232001, China.

PMID: 31247388
DOI: 10.1016/j.saa.2019.117191

Abstract

To explore a rapid detection technology for detecting heavy metals in soil based on hyperspectral data, this paper took an iron mine in Daye Country, Hubei Province, as the research area, used a FieldSpec4 portable ground object spectrometer to obtain soil spectral reflectance and combine the measured data, and used three spectral transformation methods: first-order differential, second-order differential, and continuum removal. We studied the indirect hyperspectral inversion of heavy metals in reclaimed soils in the iron mine area by using three models: partial least squares regression, support vector machine, and back propagation (BP) neural network. The results show that spectral transformation can effectively highlight the position of spectral characteristic bands and improve the correlation between spectral curves and iron (Fe) element concentration. The partial least squares regression model based on first-order differential had the highest inversion accuracy for Fe element concentration in the study area, R² and RMSE were 0.88 and 0.53, respectively. The correlation analysis of soil elements showed that the highest correlation coefficient between Cu and Fe was 0.81. We selected the copper (Cu) element with the largest correlation coefficient with the Fe element as an example and realized the indirect prediction of soil Cu concentration using a linear regression model and BP neural network model. Among them, the model based on BP neural network is better, R² was 0.82, RMSE was 0.62, compared with the direct method, the model R² increased by about 0.2, and the root mean square error decreased by about 0.1. The effect of the indirect method was better than that of the direct method. We selected the optimum statistical interpolation method for spatial analysis of Fe and Cu concentrations in the soil of the study area and further demonstrated the feasibility of the indirect inversion method of heavy metals in the soil of iron mine areas based on hyperspectral data. These results provide a theoretical basis and new ideas for the application of near-earth sensing technology in soil and for efficient detection of heavy metals in iron ore areas.

Keywords: Back propagation neural network; Hyperspectral; Partial least squares regression; Soil heavy metals; Support vector machine.

Publication types

Review