Inversion Of Soil Heavy Metal Content With KPLS Model Based On Selection Of Feature In Arable Land

Large amounts of sewage and exhaust gas produced by industrial production seriously cause the pollution of the surrounding arable land.Monitoring soil heavy metal content can effectively obtain the soil environmental pollution degree in time and accurately.Therefore,to protect limited soil resources and maintain soil environmental quality,using quantitative inversion techniques of remote sensing to invert heavy metal content of arable land soil in industrial areas is of great significance.This paper selects the soil around a industrial area in Hengdong county as the research object.we explored the potential of hyperspectral remote sensing in quantitative inversion of soil heavy metals(Cu,Pb,Hg).The spectral data is of the sample soil were resampled and combined transform.Constructing the correlation between the contents of heavy metal Cu,Pb and Hg and the transformed spectrum,and screening the optimal spectral pretreatment method corresponding to each heavy metal through qualitative analysis.Using different feature extraction algorithms to obtain the spectral characteristics of each heavy metal.The partial least squares and Kernel partial least squares regression methods were used to construct a hyperspectral quantitative inversion model of soil heavy metals Cu、Pb、Hg contents in this study area,and analysis and evaluation of the model prediction and accuracy.The main research results of this paper include:(1)Compared with the original spectrum,the spectral transformation method can improve the correlation between the heavy metal content and the spectral reflectance.The optimal combination of spectral transformation methods for Cu and Hg are Savitzky-Golay smoothing(SG)、Standard Normal Variate Correction(SNV)and first-order differential(FD),and Pb elements are Multiplication Scatter Correction(MSC)and FD,and the model established by it also has better accuracy.(2)Applying Competitive Adaptive Reweighted Sampling(CARS),Successive Projections Algorithm(SPA),and principal component analysis(PCA)to extract spectral features of Cu,Pb,and Hg elements can all reduce band redundancy and improve modeling effectiveness.According to the model inversion accuracy,the optimal feature algorithm for Cu and Hg elements is CARS,and the Pb element is SPA.(3)The partial least squares(PLS)model for Cu and Hg elements contents has better accuracy,and for Pb elements is less accurate.The kernel partial least squares(KPLS)model for Pb and Hg elements contents has better accuracy,for Cu elements is less accurate.Cu element has a strong linear relationship with the spectral characteristics,and Pb,Hg has a strong nonlinear relationship with the spectral characteristics.(4)The optimal processing paths for inversion of various heavy metal elements are different,among which Cu element is SG-SNV-FD-CARS-PLSR,Hg element is SG-SNV-FD-CARS-KPLSR,and Pb is MSC-FD-SPA-KPLSR.For different heavy metal elements,different inversion methods need to be established for rapid non-destructive testing.This study analyzes the soil spectrum through combined transformation ways and heavy metal spectrum feature extraction.According to the linear or nonlinear relationship between different heavy metals and spectrum feature,this study respectively establishes the model for analysis and evaluation.The results can play a role in the monitoring and treatment of heavy metal pollution of arable soil in industrial areas,which certainly has reference value for realizing the quantitative inversion of heavy metal content from measured hyperspectral to the application of remote sensing image to monitor pollution.