The Study Of The Flour Quality Based On Near Infrared Spectroscopy Technology

The flour is an essential raw material of food in our daily lives, therefore flour qualityhas a direct impact on the final flour processing and eating quality. We try to use the nearinfrad spectroscopy technology for the rapid determination of related components of flour,such as moisture content, protein content, ash content, starch content, dry-gluten content,wet-gluten content and gulten index. All of this work can lay the foundation for the rapidevaluation of nutritional quality and processing quality of flour.In the experimental part, GB methods were used to measure chemical values of floursamples, such as moisture content, protein content, ash content, starch content, dry-glutencontent, wet-gluten content and gulten index. The distribution range of measured moisturecontent of is10.5492-17.6423%, the distribution range of measured protein content is6.9460-19.1760%, the distribution range of measured ash content of is0.3048-2.0385%, thedistribution range of measured starch content of is58.3150-73.1450%, the distribution rangeof measured dry-gluten content of is3.33-48.2%, the distribution range of measuredwet-gluten content of is6.6-16.9%, the distribution range of measured gulten index content ofis20-100. We collect near infrared spectrum of all flour samples by using the near infraredspectroscopy instrument.In the data processing stage, we should eliminate the outliers by using the chemometricssoftware which generated by the environmental impact of the instrument causes and humanfactors. Along with the gradually increasing of the number of outliers, the predictive ability ofmodels built by using the remaining samples have a change, by comparing the models we canfind that the eliminating of appropriate outlier samples can enhance the predictive ability ofmodels. We can also find that the best outlier number of each constituent of flour is different.In order to purifying the composition informations of the near infrared spectroscopy, thespectral data should be pretreated before modeling. In order to optimizing the bestpretreatment method, this paper use orthogonal experimental method to combine thepretreatment factors, the results showed that appropriate spectral data pre-processing methodcan effectively improve the stability of the model. Comparison flour component models wecan find: the best pretreatment method of the different component model are different, themodels of the same chemical indicators which were built by using the different sample setsalso have different best pretreatment methods. Building models in different spectral region,the results showed that different ingredient has different information concentrated spectral region and all ingredients’(except drygluten) information concentrated spectral region is thewhole spectral region, therefore every kind of ingredient(except drygluten) informationdistributed in the whole spectral region. At last, this paper use five different modelingmethods (such as PCA, PLS, MPLS, LC, ANN) to build models of every constituent, theresult indicate that different ingredient has different best modeling method and propermodeling method can improve the predictive ability of the model.After above data processing, the predictive correlation coefficients of the flouringredients models (such as moisture model, protein model, ash model, starch model,drygluten model, wet-gluten model and model index model) respectively were0.905,0.807,0.956,0.833,0.845,0.859,0.474. This shows that the models except gluten-index model canbetter predict chemical constituent content of unknow flours.