Research On The Inversion Model Of Farmland Soil Heavy-metal Based On Improved Fuzzy Support Vector Machine

Under the circumstance that soil heavy metal pollution is increasingly serious,it is a trend to develop remote sensing to monitor soil heavy-metal pollution.The measured hyperspectral data has many bands and high spectra resolution.Measured hyperspectral data with multiple bands and high spectral resolution,the use of measured hyperspectra to study the inversion of heavy metal content in cultivated soil can provide the basis for image spectrum inversion of heavy metals in cultivated soil,thereby accelerating the rapid and large-scale monitoring of heavy metal pollution.Using field measurement hyperspectral prediction of soil heavy metal content through fuzzy SVM regression model is a method based on computer machine learning.However,general fuzzy support vector machines have a variety of options for optimizing the two main parameters(penalty parameter and nuclear parameter),and even some of the default choices directly,this article based on the genetic algorithm to optimize the parameters,using the optimized model to predict the content of soil heavy-metals Fe,Zn,Cu;While using partial least squares regression model to predict the content of heavy metals Fe,Zn,Cu,and was compared with the fuzzy SVM model optimized by the genetic algorithm;Researching the correlation of heavy metals in cultivated soils in Gangkou and Beishan areas,and indirect inversion of heavy metal content without significant spectral features using the correlation between heavy metals.The main findings of the paper include:(1)fuzzy support vector machine optimized by genetic algorithm has achieved better prediction results,and the average relative error of prediction is reduced by 3.16%to 7.86%.Genetic algorithm parameter optimization can improve the prediction accuracy of the fuzzy SVM regression model.(2)The results of fuzzy support vector machine model and partial least squares model optimized by genetic algorithm show that the fuzzy SVM regression model based on genetic algorithm optimization can be applied to inversion of soil heavy metal content,and the estimation result is more than partial least squares method,and it embodies the advantages of machine learning methods in mining weak information.(3)heavy metals Pb,Cd have no significant spectral characteristics,can not be directly used to predict by the regression model,but Zn-Pb,Cu-Cd has significant correlation,use its correlation indirectly predicted the contents of heavy metals Pb and Cd,with an average relative error of 14.23%?20.09%.