Quantitative Detection Of Talc In Flour Based On NIRS And Study On Variety Differences

The flour was one of the main sources of daily diet,and its safety and quality had been paid more and more attention by the society.Talcum powder was a common flour additive.However,long-term use of talcum powder could cause gastrointestinal lesions and increase the risk of ovarian cancer.Based on the advantages of near infrared spectroscopy（NIRS）in detecting material components,NIRS was used to detect the content of talcum powder in flour.This study was based on NIRS combined with chemometrics to detect the content of talcum powder in flour.In order to improve the adaptability and practicability of the detection model,123 experimental samples mixed with different contents of talcum powder were prepared based on three varieties of flour.On the basis of abnormal detection and sample division,a variety of preprocessing methods were used to optimize the spectrum.Through performance analysis,the performance of gradient boosting decision tree（GBDT）model under standard normal transformation（SNV）combined with first derivative（1D）was the best.Then,the multilevel method combining elastic network（EN）and genetic algorithm（GA）was used to select 57effective features,and the GBDT model was established on this basis.At this time,the coefficient of determination（R~2）,root mean square error（RMSEP）and relative analysis error（RPD）of GBDT model on the test set were 0.9650,1.0346 and 5.4801,respectively.Further,the adaptability of detection model on different varieties of flour samples was analyzed.Through adaptability evaluation,the adaptability of GBDT model in flour samples of variety A was relatively low.Therefore,Bayesian ridge regression（BRR）,support vector machine（SVM）and random forest（RF）models combined with residual optimization method were used to improve the adaptability of the detection model.Through the residual optimization,the adaptability of the detection model on the flour sample of variety A was improved.Among them,the RF model had the best optimization effect on the prediction residual.At this time,the R~2,RMSEP and RPD of GBDT+RF model on flour samples of A variety in the test set were 0.9625,1.198 and5.4237,respectively.The results showed that NIRS could accurately detect the content of talcum powder in flour,which put forward a new idea for the efficient and nondestructive detection of talcum powder content in flour.