Multidimensional Calibration of near Infrared Spectra for Component Concentration Measurement in Milk at Different Temperatures

A new hybrid algorithm (WPNOSC-NPLS) is proposed to eliminate the interference of temperature variation and develop the robust calibration models for determining the main components concentration in milk using the near infrared spectra. At first, the three dimensional (3D) spectra including wavelength, temperature and absorbency are constructed. Through unfolding method, the wavelet packet transform is employed to split the spectra into different frequency components, which can be refold to the corresponding 3D sub-matrices. Then, the 3D orthogonal signal correction algorithm is applied to remove the interference information irrelevant to analyte in each 3D sub-matrix. At last, the summation of these 3D sub-matrices is viewed as the filtered spectra matrix, and the N-way partial least squares (NPLS) algorithm is implemented to develop the calibration models. WPNOSC-NPLS algorithm was validated by measuring the fat concentration of milk using the NIR spectra at 25°C, 30°C, 35°C and 40°C. The results show that the WPNOSC-NPLS algorithm can effectively remove the interference of temperature variation and also can significantly improve the precision and robustness of calibration model for fat, indicating that it is a promising tool for multidimensional regression.