Temperature-perturbed two-dimensional generalized correlation characteristic slice spectra combined with multivariate...

In order to improve the discrimination accuracy of adulterated milk, a detection method was proposed based on temperature-perturbed generalized two-dimensional (2D) correlation characteristic slice spectra. A total of 240 samples were prepared including three brands of 40 pure milk and 40 urea-tainted milk, respectively. The infrared attenuated total reflection spectra of each sample were collected at different temperatures. Synchronous 2D infrared correlation spectrum of each sample was calculated under the external perturbation of temperature. The characteristic slice spectra of each sample were extracted from synchronous 2D correlation spectrum at characteristic peaks of milk and adulterants. N-way partial least squares discriminant analysis (NPLS-DA) models of single brand and the fusion of three brands of adulterated milk were established based on 2D correlation characteristics slice spectra. For comparison, the discrimination models were established using synchronous 2D correlation spectra and one-dimensional (1D) infrared spectra at room temperature, respectively. For the three brand fusion models, the discrimination accuracies of unknown samples were 100%, 98.8% and 82.7% using 2D correlation characteristic slice spectra, 2D correlation spectra, and 1D spectra, respectively. The results showed that the proposed method not only compressed the data, but also effectively extracted the characteristic information, and improved the accuracy of discrimination.