Vis-NIR spectroscopy Combined with Wavelengths Selection by PSO Optimization Algorithm for Simultaneous Determination of Four Quality Parameters and Classification of Soy Sauce

The performance of Vis-NIR techniques combined with variable select by a simple modified particle swarm optimization (PSO) algorithm for the determination of four quality parameters in soy sauce was evaluated. Compared with full-spectral support vector machine regression (Full-SVMR) and SVMR based on competitive adaptive reweighted sampling (CARS-SVM) method, the application of PSO wavelength selection provided a notably improved SVM regression model. The root-mean-square error of amino acid nitrogen, salt, total acid content, and color ratio obtained by PSO-SVMR are 0.0075 g/100 ml, 0.2176 g/100 ml, 0.0077 g/100 ml, and 0.0506 in predicted sets, respectively. The correlation coefficients of predicted sets obtained by PSO-SVMR reached 0.9997, 0.9462, 0.9996, and 0.9998, respectively. Meanwhile, a classification study constructed with principal component analysis and SVM classification model based on the feature wavelengths selected by PSO shows that Vis-NIR spectra can be used to classify soy sauce according to their brands and quality. The result showed that the Vis-NIR spectroscopy technique based on PSO wavelength selection has high potential to predict the quality parameters in a nondestructive way. This analytical tool may also contribute to the detection of fraud and mislabeling in the soy sauce market and certainly contribute to improvement in quality and reliability of the soy sauce market.