GMDH Model Inversion Of Soil Chromium Content In Cultivated Land Under Pixel Linear Unmixing

The increasingly serious soil heavy metal pollution has become the focus of world attention.How to use remote sensing to realize the rapid monitoring of soil heavy metal content has been concerned by scholars all over the world.This article selects the cultivated soil in a certain area of Hengdong County as the research object,selects the appropriate remote sensing image data source,and combines the measured heavy metal detection data.The spectral linear unmixing was carried out by dimidiate pixel model,and the response relationship between the spectral data after the linear unmixing and the original spectral data of remote sensing image to the heavy metal elements in soil was discussed,as well as the response relationship between the spectral reflectance and the heavy metal elements in soil under different spectral transformation processing.Based on this,the GMDH(Group Method of Data Handling,GMDH)estimation model of soil reflectance and heavy metal Cr content under different spectral transformations was studied and constructed.The main research results of this paper are as follows:(1)The cultivated soil in the study area is suffering from the pollution stress of heavy metal elements,and the spatial variability of the heavy metal elements in this order is:Cd>Pb>Zn>Hg>Cu>Crr,among which the variation degree of Cr is the lowest,and the spatial distribution difference in the study area is small;(2)Dimidiate pixel linear unmixing of remote sensing images is used to extract the image soil reflectance,and compares the correlation with the heavy metal Cr between the original image reflectance and the image soil reflectance under different spectral transformations and heavy metal Cr.The results show that the number of significant correlation bands between the unmixed spectra and heavy metals increases and the correlation increases;(3)The PLS model and GMDH neural network model are established for the image original reflectance and the image soil reflectance after the reciprocal logarithm transformation.research results show that the GMDH model constructed based on the spectral reflectance and heavy metal Cr content after demixing has better stability,model prediction ability and accuracy evaluation effect,and better model generalization ability.In this study,through linear unmixing of remote sensing images to obtain the image soil reflectance and the establishment of a quantitative inversion model of soil heavy metal Cr content in the study area can provide a useful reference for the large-scale monitoring of soil heavy metal pollution.