Differentiation of Soy Sauce by Pattern Recognition Analysis of Mid- and Near-IR Spectra

Abstract For certificating the authenticity of soy sauce, mid- and near-IR spectra of 27 commercial soy sauce produced from whole soybeans and 30 from defatted soybeans were measured. Factor analysis applied to MIR and NIR spectra individually indicated the existence of some difference between two soy sauces. Then, SIMCA, conventional step-wise linear discriminant analysis (LDA) and LDA using genetic algorithms (GALDA) were applied to their spectra for classifying soy sauce samples according to types of soybeans. In step-wise LDA, 94.7% of samples were correctly assigned based on six wavenumbers in mid-IR spectra but 100% correct classification was obtained by six wavelengths in NIR spectra. Cooman's plots in SIMCA applied to mid- or near-IR spectra indicated that differentiating two types of soy sauce samples was difficult due to high similarity in their constituents. Perfectly correct classification was attained by five to eight mid-IR wavenumbers and four to five near-IR wavelengths selected by GALDA. In general, the differentiation using NIR spectra was more reliable than that using MIR spectra.