Abstract
Meat, as an important source of protein, is one of the main parts of many people's diet. Due to economic interests and thereupon adulteration, there are special concerns on its accurate labeling. In this study Fourier transform infrared (ATR-FTIR) spectroscopy combined with chemometric techniques (principal component analysis (PCA), artificial neural networks (ANNs), and partial least square regression (PLS-R)) were employed for discrimination of pure beef meat from textured soy protein plus detection and quantification of texture soy protein in a mixture with beef meat. Spectral preprocessing was carried out on each spectra including Savitzki-Golay (SG) smoothing filter, Standard Normal Vitiate (SNV), scatter correction (MSC), and min-max normalization. Spectral range 1700-1071 cm(-1) was selected for further analysis. Principal component analysis showed discrete clustering of pure samples. In the next step, supervised artificial neural networks (ANNs) were performed for classification and discrimination. The results showed classification accuracy of 100% using this model. Furthermore, PLS-R model correlated the actual and FTIR estimated values of texture soy protein in beef meat mixture with coefficient of determination (R(2)) of 0.976. In conclusion, it was demonstrated that ATR-FTIR spectroscopy along with PCA and ANNs analysis might potentially replace traditional laborious and time-consuming analytical techniques to detect adulteration in beef meat as a rapid, low cost, and highly accurate method.