Research On Modeling Of Naked Oatmeal Quality Detection Method Based On Machine Learning

The purity of naked oats flour is the key index to determine its quality.Naked oats flour sold on the market will be mixed with corn starch or potato starch to varying degrees to adjust the cost of naked oats flour.There is no good solution to how to effectively and quickly identify naked oats.Near infrared spectroscopy(NIRS)is a technology which can analyze the spectral characteristics of substances quantitatively and qualitatively,and has the advantages of non-contact,multi-component prediction,low cost and so on.In this study,near infrared spectroscopy technology is applied to the discrimination analysis of naked oats grade and naked oats to provide a nondestructive testing method.Taking naked oats,corn starch and potato starch as the research objects,the spectral data were collected by Fourier near infrared spectrometer.According to the technical process of spectral analysis,the classification and discrimination model of naked oats,corn flour and potato starch was established,the classification and discrimination model of naked oats mixed with different amounts of corn starch was established,and the classification and discrimination model of naked oats mixed with different amounts of potato starch was established.The main work is as follows:(1)In spectral preprocessing,S-G(savitzky Golay,S-G)smoothing,MSc(multiplicative scatter correction),derivative and normalization preprocessing methods are used to denoise the original spectrum,and the spectral preprocessing method is determined based on the minimum principal component score.The specific processing steps are as follows:the original spectra of naked oat flour,corn starch and potato starch are pretreated with first-order derivative,second-order derivative and normalization respectively.Through the comparative analysis of the processing results,which group of data is used for the next step is determined.The data of naked oat flour mixed with corn starch and potato starch were processed in the same way.The results were analyzed and compared to determine the pretreatment method of this group of data.(2)In the principal component analysis,the preprocessed data is analyzed by principal component analysis to obtain the principal component analysis diagram.The principal component score that should be selected in the modeling process is determined based on the standard that the cumulative contribution rate of the principal component reaches more than 90%.The selected principal components are used for the following modeling analysis.(3)In the modeling analysis,the selected principal components are used as classification variables,BP neural network and support vector machine are used for modeling respectively,and the optimal discrimination model is determined based on the test accuracy.In the classification modeling of naked oat flour,corn starch and potato starch,by comparing the accuracy of the confusion matrix obtained by BP neural network and support vector machine,it is determined that the optimal discrimination model is the support vector machine under the first derivative;In the classification modeling of naked oat flour mixed with corn starch,through the comparison and analysis of its confusion matrix,it is determined that the optimal classification model is the support vector machine under the first derivative;In the classification model of naked oat flour mixed with potato starch,by determining the test accuracy,it is determined that the best discrimination model is BP neural network classification under the first derivative.