Research On Hyperspectral Remote Sensing Monitoring Method Of Heavy Metals In Black Soil Region

The rapid collection,accurate analysis and comprehensive evaluation of heavy metal pollution data in farmland soil are extremely important for the growth environment and safety of agricultural products in black soil area.However,heavy metal monitoring by means of sample analysis is time-consuming and laborious with low timeliness.Nowadays,with the development of hyperspectral remote sensing technology,inversion mapping through soil spectral information has become a research hotspot,and a breakthrough has been found for the large-scale dynamic monitoring of heavy metal pollution elements in farmland black soil.This paper took the typical black soil area in the northeast of Hailun City,Heilongjiang Province as the research object.The chemical detection data of samples,ASD ground hyperspectral data and CASI/SASI aviation hyperspectral data were used to analyze the occurrence relationship of soil components and the spectral characteristics of soil,and to indirectly establish an accurate model for inversion of As,Zn and Cr content.Finally,the estimation mapping of Cd,As,Hg,Cr,Cu,Pb,Ni and Zn contents in soil polluted by heavy metals was completed.By analyzing the Pearson coefficient,principal component analysis(PCA)and cluster analysis(CA)among the chemical components of black soil,it was found that the content of As,Zn and Cd were positively correlated with the concentration of iron oxide.Based on the above conclusion,an indirect prediction model of As,Zn and Cd content with Fe2O3 as the intermediate quantity was established.The direct inversion model based on multiple stepwise regression(MSR),partial least squares(PLSR),back propagation neural network(BPNN).first-order differential(FD),continuous spectrum removal(CR)and multiple scattering correction(MSC)transformations was compared with the prediction accuracy of the model.It was found that the R~2 and RPIQ of the indirect prediction model was greater than those of the direct inversion model,indicating that the indirect prediction model was more accurate reliability.After resampling according to the ground spectrum,the Airborne Hyperspectral spectrum was brought into the direct inversion and indirect prediction model of As,Zn and Cd contents based on the ground spectrum.It was found that the inversion accuracy was lower than the prediction accuracy based on aerial hyperspectral feature,and the model could not be transferred directly.Then,the first order differential(FD),second order differential(SD),third order differential(TD),and continuous spectrum removal(CR)feature bands of aviation hyperspectral were extracted by using the correlation coefficient method,and the optimal inversion mapping process for each heavy metal polluting element in aviation hyperspectral was found in different modeling methods:As_CR_SVM,Ni_CR_SVM,Cr_TD_PLSR,Hg_TD_BPNN,Zn_CR_SVM,Cd_TD_PLSR,Cu_TD_PLSR,Pb_TD_PLSR.The determination coefficient R~2 comparison in the order:0.404,0.328,0.321,0.284,0.281,0.267,0.208,0.092.The aerial hyperspectral inversion results had good spatial coincidence with the known heavy metal element content distribution data,demonstrating the application potential of aerial hyperspectral for large-scale rapid visual monitoring of black soil heavy metal contaminated elements.