Quantitative Estimating Of Soil Heavy Metals With Hyper-spectrum In Polymetallic Mining Areas

Studying the soil heavy metals in Nianzigou polymetallic mine, Pingquan, Hebei,this paper explored the potential of hyperspectral remote sensing in quantitativeinversion of soil heavy metals (As, Cr, Hg, Pb) and the impact of extraction ofspectral features in preprocessing. Analyzed and compared the effectiveness ofmultiple linear regression, partial least squares regression and least squares-supportvector machines methods for quantitative estimating of soil heavy metals withhyper-spectrum, and the results showed that the least squares-support vectormachines inversion method was the best methord. The main contents and results areas follows:(1) Moving average, Savitzky-Golay, median filter and Gaussian filter methodwas applied to preprocess the original soil hyperspectral data, and evaluated thesmoothing effects. The results showed that the effect of Savitzky-Golay was the best.On the basis of spactral preprocessing, employed the methods of continuum removal,first/second derivative, standard normal variable transformation and logarithmic totransform spectral data. The results showed that spectral transformed method haveenhancement effect on the spectral characteristics.(2) Respectively, based on continuum removel, correlation analysis and principalcompontent analysis method to extract soil spectral characteristics, and thenoptimized the features as the input arguments for the subsequent estimation models.The results showed that: the characteristics extract based on continuum removel andcorrelation analysis method can be a good characterization of heavy metals in soilspectrum characteristics, the characterization of characteristics extract based onprincipal compontent analysis method is poor. The number of characteristic variablesselected for elements As, Cr, Hg, Pb are10,10,9,4. And absolute values of thecorrelation coefficients were no smaller than0.77,0.60,0.43,0.49.(3) The methods of multiple linear regression, partial least squares regressionand least squares-support vector machines were applied to establish the relationshipbetween spectral characteristics and heavy metals concentrations. It was used the R2to evaluate the stability of model, and RMSEC and RMSEP to compute the accuracyof model. The results showed that the stability and accuracy of model established bythese three methods was good. The least squares-support vector machines performed best, the R2of modeling and prediction model of As, Cr, Hg, Pb achieved0.93,0.92,0.81,0.57and0.84,0.68,0.68,0.37, respectively.