Research On Near Infrared Spectroscopy Of Wheat Quality Based On Model Population Analysis

As an important food,wheat is rich in some nutrients needed by human body.It can be used to process the daily necessities of adults,such as bread,noodles,biscuits,etc.The content of moisture and protein are important indexes to evaluate the quality of wheat.The method of non-destructive rapid determination of wheat component content is of great significance to the inspection and processing of wheat and other agricultural crops.At present,near-infrared nondestructive testing technology is based on one-time modeling.Although the modeling method is simple,the related prediction results are not accurate because of "Small sample,large variable".The key problem hindering the further development and application of this technology is the accuracy and stability of the prediction model.In this paper,based on this background,the near infrared spectrum analysis method of wheat quality was studied,the main research contents are as follows:Firstly,wheat samples from different regions of the country were collected,and the chemical values of water and protein were tested respectively.Meanwhile,near-infrared spectral data of wheat samples were collected by optical fiber spectrometer.Secondly,the NIR data of the original wheat was preprocessed based on normalization and abnormal sample diagnosis and elimination methods to optimize the spectral data of the wheat samples.Finally,According to model population analysis(MPA),the characteristic variable was extracted which is combined with bootstrapping soft shrinkage(BOSS),competitive adaptive reweighted sampling(CARS),Automatic weighting variable combination population analysis(AWVCPA),iteratively retains informative variables(IRIV),Monte Carlo variable combination population analysis(MCVCPA),variable combination population analysis(VCPA),variable combination population analysis-iteratively retains informative variables(VCPA-IRIV),and the prediction model of wheat moisture and protein content was established.The result shows that the root mean-squared error of prediction(RMSEP)by BOSS-PLS is reduced from 0.4717 to 0.2072 in wheat moisture content prediction;the root mean-squared error of prediction RMSEP by BOSS-PLS is reduced from 0.4936 to 0.2105 in wheat protein content prediction,which is compared with full spectrum model.The prediction accuracy has been improved by 56% and 57.3% respectively.The model has been greatly simplified and the prediction ability has been greatly improved.