Inversion Of Soil Cu Content Based On WorldView-3 Data

Soil is an invaluable resource and monitoring of heavy metal pollution in the soil is essential.The traditional geochemical method is time-consuming and laborious,and the use of remote sensing images to invert the heavy metal content of the soil greatly reduces the monitoring cost and improves the efficiency.In this paper,the WorldView-3 multi-spectral high-resolution remote sensing image is used to invert the soil heavy metal Cu content in Dexing Copper Mine.The research aims to improve the inversion precision from the aspects of input data preparation,empirical model selection and accuracy evaluation index.This paper has achieved the following results:1.The average Cu content of the sample in the study area is 1006 mg/kg,which is 20 times of the background value,indicating that the degree of Cu contamination in the study area is very serious.On the selected soil samples,the abnormal samples were excluded according to the Cu content by the box plot method,and then the soil samples were removed from the image,and 110 sets of available soil samples were obtained as Cu content input data.2.Due to the high spatial resolution,wide spectral range and large coverage of WorldView-3 image data,this paper selects the soil Cu content in the WorldView-3image inversion study area.The extracted soil reflectivity is subjected to reciprocal logarithm and first-order differential processing,and a mixed spectral variable composed of a plurality of morphologically correlated bands is newly added in the input spectral variable,which is simply referred to as a mixed spectrum.Experiments show that the prediction accuracy based on hybrid spectral modeling is higher than that of any other single morphology.3.In the selection of empirical models,combined with the original spectrum of the soil,the inverse logarithmic spectrum,the first-order differential spectrum,the mixed spectrum and the soil Cu content to establish partial least squares,support vector machine,BP neural network,random forest inversion Model,four modeling methods for the Cu content prediction results of the validation set determination coefficientR_p~2 because the input spectral variables are different,so to combine the performance of the two evaluation indicatorsMRE_p andRMSE_p,we can see that the hybrid spectral model built by BP neural network The performance is the best,so the model is selected to plot the spatial distribution map of Cu content in the study area,and the obtained thematic map is compared with related data.It is found that there is a certain error in the content distribution of heavy metal Cu,but the overall distribution prediction is reasonable.