Abstract: The models of two brands of adulterated milk were constructed based on the information fusion of near and mid infrared spectroscopy combined with partial least squares discriminant analysis（PLS-DA）. 40 Yili pure milk samples, 40 Mengniu pure milk samples, 40 Yili milk adulterated with melamine samples, and 40 Mengniu milk adulterated with melamine samples, were prepared respectively. Then, near infrared transmittance spectra and mid infrared attenuated total reflectance spectra of all samples were obtained under room temperature. Based on the fused data of near and mid infrared spectra, the discrimination models of Yili, Mengniu and both brands milk were established. The classification accuracies of three discrimination models for prediction set were 96.2%, 96.2% and 94.2%, respectively. For comparing, the discriminant models of Yili, Mengniu and two brands milk were established using near infrared spectra and mid infrared spectra individually. The results show that the fusion of near and mid infrared spectra not only can reduce the effect of different brands milk on the model, but also can provide better results compared with near infrared spectra and mid infrared spectra individually.

Key words: near infrared spectroscopy, mid infrared correlation spectroscopy, data fusion, different brands milk, discriminant analysis, melamine