| As one of the most important grain crops in China,the quality of the wheat field has a direct impact on the production of winter wheat.Soil total nitrogen,as an important nutrient of wheat field soil,is an important index to measure the soil fertility characteristics of wheat field,and also an essential element for the growth and development,yield and quality formation of winter wheat.The traditional determination of total nitrogen in soil is mainly through laboratory chemical analysis.The results are reliable and accurate,but there are many problems such as long time consuming,cumbersome operation,high cost and environmental pollution.Hyperspectral analysis technology,as a new modern science and technology,has the characteristics of fast,low cost,environmental protection,no damage and repeatability.It can acquire field data in large area and timely,and meet the development demand of precision agriculture.In this experiment,Wenxi County Soil as the research object,based on multiple samples and the total nitrogen content of soil spectral data acquisition in the county,of different size of soil with high spectral response mechanism,extraction and determination of Hyperspectral Feature Information of soil nitrogen,and comparative analysis of different soil particle size,spectral pretreatment methods and the model construction method of influence on soil total nitrogen model of overall performance,and determine the realization of spectral pretreatment methods of soil nitrogen monitoring the best diameter of soil particles,finally realize the accurate spectroscopic monitoring of the content of total nitrogen in soil.The main conclusions are as follows(1)Have obvious effects of soil particle size on soil spectral reflectance curves,and the scope of the whole band,changes of different size of soil spectral reflectance is consistent with the decrease of soil particle size,soil spectral reflectance showed a different rate of increase,soil particle size and soil spectra there was a negative correlation between the reflectance.(2)The total nitrogen content of soil is positively correlated with the original spectral reflectance of soil.When the reflectivity is processed by different transformation,the fluctuation of correlation coefficient fluctuates greatly,and the correlation has been improved obviously.The correlation between FD pretreatment spectrum and soil particle size and total nitrogen content was the best,and the absolute values of correlation coefficients were 0.973 and 0.579 respectively.(3)In this paper,we use stepwise multiple linear regression to extract the sensitive band of soil total nitrogen,which is mainly located near the visible region near 410nm,and near infrared band near 1087nm,1330nm-1750nm and 2030nm(4)The establishment of soil nitrogen PLSR linear different particle size of full spectrum SVM model,nonlinear full spectrum model and PLS-SMLR feature band model,compared with PLSR linear model and nonlinear full spectrum SVM full spectrum SVM model shows that the nonlinear model of optimum performance,to achieve better prediction results;comparison of nonlinear SVM the full spectrum model and PLS-SMLR feature band model shows that the nonlinear model of SVM model to construct different size soil nitrogen content model of the overall performance is superior to the characteristic band based on PLS-SMLR.Among them,the soil total nitrogen monitoring model based on FD transform spectrum of 0.25mm particle size and SVM model is the best.The accuracy of calibration model is the best(R2=0.972),the accuracy of validation model is the highest(R2=0.971),and the verification error is the smallest(RMSEP=0.026).The total nitrogen content model constructed in this study has a good prediction ability,which can be used to estimate the total nitrogen content accurately and quantitatively.The PLSR-SMLR model is based on the extraction of the characteristic band.The model is simple and easy to calculate,and achieves the accuracy of statistical prediction.It has certain application potential. |
PDF Full Text Request