Prediction Of Soil Properties Based On Hyperspectral Characteristics

Soil as the foundation of agricultural production,it is particularly important to obtain continuous and reliable information of soil property and its spatial distribution for precision agriculture.The routine chemical analysis method for determination of soil property content has long cycle and high cost,which cannot meet the requirement of fast and effective monitoring of soil property.In recent years,hyperspectral analysis technology has provided an effective way to obtain soil property information,which is fast,simple and efficient.In this study,soil samples were collected from Mengcheng and Jiangyan,and the spectral reflectance curves and soil organic matter content of the samples were determined.First,the mathematical transformations such as first-order differential,inversion-logarithm,and continuum removal were performed on spectral reflectance,and the soil organic matter content prediction model based on soil type and spectral classification was studied,and the superior spectral transformation method was selected.Then,some linear regression models was established based on the feature indexes,and different selection methods were selected to choose the feature bands,and principal component regression model and partial least squares regression models were established respectively.Finally,the model accuracy was evaluated and analyzed,and the influence of different methods on the accuracy of soil organic matter prediction model was discussed.It provides methodological support for rapid and accurate prediction of soil organic matter,and provides research basis for hyperspectral remote sensing mapping of soil organic matter,and provides basic ideas for further promoting the development of precision agriculture,which has great practical and scientific significance.The main conclusions are as follows:(1)The spectral transformations such as first-order differential,inversion-logarithm,and continuum removal were performed on the original spectral data,and spectral similarity was used to classify the total samples in the study area.Based on the total samples,different soil types and spectral classification,the different soil organic matter prediction models were established and the model accuracy was improved after spectral classification.(2)The spectral characteristic indexes was constructed based on the spectral data of original spectrum,inversion-logarithm and continuum removal in the study area,respectively.The characteristic indexes include high correlation band,the curvature of the bow,difference index(DI),ratio index(RI)and normalized difference index(NDI),which established some linear regression models with the characteristic indexes as independent variables.The accuracy of the model based on DIS RI and NDI was higher than that of the high correlation band and the curvature of the bow,and the prediction accuracy of soil organic matter was also improved.(3)The characteristic bands selection of original spectrum,inversion-logarithm spectrum and continuum removal spectrum was based on successive projections algorithm(SPA),uninformative variables elimination(UVE)and competitive adaptive reweighted sampling(CARS),and established the PCR and PLSR model of soil organic matter.By evaluating the accuracy of the model,it was concluded that the accuracy of the model was generally improved after characteristic bands selection,and the model with the characteristic band selection using the CARS method had the highest accuracy.At the same time,after using the characteristic bands selection method,the number of variables used to construct the model was reduced,and the model efficiency was also improved.This study provides a theoretical basis for the rapid prediction of soil organic matter.Figure[33]table[14]reference[97]...