Near Infrared Prediction Model Based On Applying The Phosphorus, Potassium Fertilizer Content Of Corn

In this study, the corn as the research object, respectively in2013to phosphorus, potassium stress test design, were set in five trophic levels, each level set three repeat. In Maize Jointing Stage by using FieldSpec3Portable NIR. spectrometer with blade detector, spectral data acquisition of corn leaf, three parts. Application of nine kinds of pretreatment methods on the original spectral information pre-processing, by comparing the optimal pretreatment method. We extract effective characteristic wavelength and principal components, using a variety of modeling method to establish the mathematical model of phosphorus, potassium, and gain the forecasting result of ideal. The main research contents and conclusions are as follows:(1) The3point smoothing,5point smoothing,7point smoothing, SNV, MSC, first derivative, two order differential, differential+SNV, two order differential+MSC nine pretreatment method for spectral data pretreatment, by comparative analysis,3point smoothing preprocessing method is the best method. Potassium under calibration set corn spectral data after3point smoothing the correlation coefficient R is0.944227, the root mean square error of RMSEC is0.333984; the prediction set of spectral data by pretreatment of the correlation coefficient is0.981556, the root mean square error of RMSEP is0.192061. Single phosphorus under calibration set corn spectral data and the correlation coefficient is0.904486, the root mean square error of RMSEC is0.437067; the prediction set corn spectral data by pretreatment of the correlation coefficient is0.955965, the root mean square error of RMSEP is0.296765. Therefore,3point smoothing is optimal pretreatment method.(2) The corn phosphorus3point smoothing of spectral data using full band and partial least squares (PLS) and support vector machine (SVM) prediction model, the prediction accuracy was96%and88%respectively.The corn phosphorus3point smoothing of spectral data using continuous projection to extract11characteristic wavelength were378nm,381nm,388nm,515nm,550nm,654nm,688nm,709nm,1903mn,2089nm,2500nm, the spectral information is reduced, respectively. A predictive model using SPA-PLS and the SPA-SVM modeling method, the accuracy of prediction were84%and80%, based on model simplification received better forecasting results. The spectral data of3point smoothing after using MATLAB software to extract6principal components, simplifying the spectral bands, using BP neural network prediction model, the accuracy rate of forecast is84%, achieved good prediction accuracy. The characteristic wavelength of modeling and component modeling in a certain extent, reduced the workload, but its precision is less than full band model, so the full band PLS modeling modeling methods for the best.(3)The potassium in maize3point smoothing of spectral data using full band and partial least squares (PLS) and support vector machine (SVM) prediction model, the prediction accuracy was100%and92%respectively.The potassium Corn3point smoothing of spectral data using continuous projection method to extract7features wavelength were355nm,357nm,360nm,378nm,1000nm,1405nm,2010nm, respectively, the spectral information is reduced, respectively SPA-PLS and SPA-SVM modeling method to establish the prediction model, the the prediction accuracy were88%and84%, received a relatively good results. After the preprocessing of the data using MATLAB software to extract8principal components, the spectral segment has been reduced, by using BP neural network prediction model, the accuracy rate of forecast is88%, achieved good prediction accuracy. By comparison, the full band PLS modeling modeling methods for the best.Taking maize as carrier,Spectral characteristics of phosphorus, potassium stress using near infrared spectroscopy are presented and achieved better results.Lay a solid foundation for further study on the stress of maize nutrition.