Rapid Detection Of Additives And Illegal Additives In Wheat Flour Based On Hyperspectral Technology

A large supply and demand of wheat flour,intense processing industry competition,and consumers’diversified factors cause illegal and excessive problems of wheat flour additives.The routine detection method cannot have satisfied the market volume,fast,non-destructive and accurate detection requirements.It is necessary to develop a detection technology with batch testing,good reproducibility.This experiment explored the feasibility of high-efficiency nondestructive detection of additives（azoformamide）and illegal additives（talcum powder and benzoyl peroxide）in wheat flour by hyperspectral technology in the range of900～1700 nm.The specific results are as follows:（1）Rapid nondestructive detection of talcum content in wheat flour based on hyperspectral technology.The original spectral data of R-mode of mixing samples were collected and converted into A-mode spectral data and K-M mode spectral data respectively by matalb data analysis software.SNV,MSC,NOR,BC were used to preprocess the three original spectra.PLS prediction models were established.By comparing the performance parameters of the full-band models,it was found that the optimal spectral mode and pretreatment of talcum content prediction model were A mode and BC pretreatment respectively.In order to improve the efficiency of model operation,PLS model and MLR model were established by using RC,SPA and CARS feature wavelength screening methods.Compared with the performance of full-band prediction model,it can be known that the best prediction model was A-BC-RC-MLR.r_P is 0.992,RMSEP is 1.784%,RPD is 8.251.（2）A rapid nondestructive prediction of BPO content in wheat flour was performed based on hyperspectral technique combined with multiple stoichiometric algorithms.Hyperspectral imaging system collected original spectral data of samples in R mode,and obtained spectral data of A mode and K-M mode respectively through transformation of spectral mode.The prediction models were developed by using GFS,SGS,MSC and BC.The best spectral mode and pretreatment method for full-band model were R mode and GFS pretreatment respectively.For simplifying the operation,the characteristic wavelengths were selected by using three wavelength screening methods（RC,SPA,CARS）.The results showed that the best prediction model was R-GFS-RC-PLS,with r_P of 0.989,RMSEP of 28.087 mg/kg,RPD of 6.269,respectively.（3）The content of ADA in wheat flour was rapidly detected by hyperspectral technology and chemometrics.After obtaining the hyperspectral image data of R mode and converting it into the original spectral data of A mode and K-M mode,the pretreatment methods of NOR,GFS,MSC and BC were used to eliminate the interference factors of instrument and environment.PLS prediction models were constructed based on three spectral model respectively,and K-M mode was found to be the best spectral mode.The best model was built based on raw data.In order to reduce the problem of high-dimensional collinearity,RC,SPA and CARS were used to select the characteristic wavelengths and simplify the prediction model.Combined with PLS and MLR algorithm,the full-band prediction model was optimized.The results showed that the best model for predicting ADA content in wheat flour was KM-RAW-SPA-MLR and r_P was 0.996,RMSEP and RPD were 14.727 mg/kg and 11.955,respectively.