COMPARISON OF PRINCIPAL COMPONENT REGRESSION (PCR) AND PARTIAL LEAST SQUARE (PLS) METHODS IN PREDICTION OF RAW MILK COMPOSITION BY VIS-NIR SPECTROMETRY. APPLICATION TO DEVELOPMENT OF ON-LINE SENSORS FOR FAT, PROTEIN AND LACTOSE CONTENTS

摘要

Visible and Near InfraRed (VIS-NIR) spectrometry from 400 to 1100 nm in addition to Partial Least Squares (PLS) regression or Principal Component Regression (PCR) is a very interesting method to measure several important parameters of non-homogenised fresh milk such as fat, lactose and total protein content. These parameters can be used to analyze the nutritional properties of milk and, consequently they are very important to determine the economic value of produced milk. This paper studies and compares the potential use of PCR and PLS statistical methods to obtain the values of milk nutrients composition in milk, and present the application to the development of on-line sensors for those nutrients. The potential of VIS-NIR spectrometry in a spectral region below 1100 nm has been studied in this paper due to working in this region, a low-cost system would be obtain. Several fresh milk samples taken during milking process were analyzed by means of standard measurement procedures and VIS-NIR spectrometry in order to verify the capabilities and precision of proposed method. As will be seen in next sections, this method is very interesting for fat content estimation, but it present some problems for total protein and lactose measurement, probably due to the low value of protein and lactose spans.