Quantitative analysis of melamine by multi-way partial least squares model with two-dimensional near-infrared correlation spectroscopy

A new approach for quantitative analysis of melamine in milk was proposed based on two-dimensional (2D) correlation near-infrared spectroscopy and multi-way partial least squares (N-PLS) in this paper. 40 pure milk samples and 40 milk samples adulterated with different contents of melamine were prepared. The near-infrared transmittance spectra of all samples were measured at room temperature. Then 2D NIR-NIR correlation spectroscopy under the perturbation of adulterant concentration was calculated and N-PLS model for the melamine concentration was established with 2D correlation spectra (28x51x51). For the prediction set, the root mean square errors of prediction (RMSEP) for melamine concentration was 0.067 g/ L and the coefficient correlation between actual reference values and predicted values was 0.999, which means the model has good predictive ability. For comparison purpose, partial least squares (PLS) model was also built using the conventional one-dimensional near-infrared spectra (28x51), where the RMSEP and the coefficient correlation were 0.079 g/ L and 0.998, respectively. The average relative prediction error was 22.9% for N-PLS model; whereas it was 122.4% for PLS model. The N-PLS models yielded relatively low RMSEP and average relative prediction error as compared to PLS model. Therefore, N-PLS method was more robust than PLS method for accurate quantification of the concentration of melamine in milk.