[Study on noninvasive detection using NIR diffuse reflectance spectrum for monitoring protein content in milk powder]

A quick and nondestructive method using near-infrared (NIR) diffuse reflectance spectroscopy to determine the content of protein in milk powder was presented. Twenty-nine kinds of milk powder were selected from domestic different districts. Partial least square (PLS) regression was utilized to build the calibration between the spectra and the protein content with chemical analysis as a reference method. Root mean square error of prediction (RMSEP) of 0. 687% was acquired. With the application of the method of wavelength selection, the number of wavelengths was decreased greatly from 1 558 to 11. The speed was accelerated and the accuracy of the calibration was improved to some extent. In order to reduce the particle size and nonspecific scattering effect, standard normal variate (SNV) and multiplicative scatter correction (MSC) were performed. These two treatments proved useful, and the RMSEP was reduced by 37%. Our results suggested that NIR diffuse reflectance spectroscopy was fast and real-time and could meet the practical measurement accuracy as required for industrial process analysis, providing a new tool for quality control of milk powder products.