[Study on quality detection of milk powder based on near infrared spectroscopy (NIR)].

The traditional NIR model was usually built according to various parameters of an individual type of milk powder so that it's really time-consuming. To simplify the application of NIR in real-time quality detection of milk powder, it was proposed in the present paper to build NIR models for a sample set composed of different types of milk powder. With 70 samples provided by one manufacturer, 6 NIR models including acidity, fat, lactose, sucrose, protein and ash, were built by optimizing algorithms. The results indicated that these NIR models except the acidity model have good stability and high prediction ability (RSD<10%, RPD>3).