application of ft-ir spectroscopy with various classification and regression models for detection and quantification of sodium hydrosulfite in iranian wheat flour

Wheat flour is one of the most important and strategic food resources especially in developing countries. the addition of sodium hydrosulfite to flour for improving some appearance features can have dangerous impacts on the consumer health. therefore, detection of this harmful substance is great practical significance. in the present study, the potential of fourier transform-mid infrared (ft-mir) spectroscopy in 400-4000 cm-1 for the fast detection of sodium hydrosulfite powder in wheat flour was investigated. after getting the spectral data from samples, firstly some preprocessing methods were used to correct harmful and unwanted effects on spectral data, and then principal component analysis (pca) as unsupervised and support vector machine (svm) and artificial neural network (ann) models as supervised classification models and partial least square regression (plsr) as regression model were applied to detect and quantify the adulteration in pure flour samples. the best outcomes were the accuracy of 86.66 and 86.70 for svm and ann models with s-g + d2 + snv preprocessing, respectively and r2p = 0.99 for plsr model.

application of ft-ir spectroscopy with various classification and regression models for detection and quantification of sodium hydrosulfite in iranian wheat flour

wheat flour is one of the most important and strategic food resources especially in developing countries. the addition of sodium hydrosulfite to flour for improving some appearance features can have dangerous impacts on the consumer health. therefore, detection of this harmful substance is great practical significance. in the present study, the potential of fourier transform-mid infrared (ft-mir) spectroscopy in 400-4000 cm-1 for the fast detection of sodium hydrosulfite powder in wheat flour was investigated. after getting the spectral data from samples, firstly some preprocessing methods were used to correct harmful and unwanted effects on spectral data, and then principal component analysis (pca) as unsupervised and support vector machine (svm) and artificial neural network (ann) models as supervised classification models and partial least square regression (plsr) as regression model were applied to detect and quantify the adulteration in pure flour samples. the best outcomes were the accuracy of 86.66 and 86.70 for svm and ann models with s-g + d2 + snv preprocessing, respectively and r2p = 0.99 for plsr model.