Rapid Detection Of Dairy Ingredients Using Near Infrared Spectroscopy

Near Infrared Spectroscopy (NIRS) has gained focuses by researchers world-wide.Since NIRS has the feature of security, preciseness and efficiency, it is effectively appliedto the analysis and testing of dairy ingredients. NIRS has developed rapidly in foreigncountries and many companies are already using NIRS online testing systems to monitorproduction.In the thesis, we collect the samples of milk powder, buttermilk, whey powder andcondensed milk, combing with the chemometric methods to create the calibration modelsfor a near infrared spectrometer L’InfraXact (FOSS，Denmark), so as to achieve real-timemonitoring in the dairy production.First, we analyze various parameters of dairy products by standard methods in thelaboratory, and then we create the calibrations models by using the methods of Partial LeastSquares (PLS) and Artificial Neural Network (ANN). The evaluation of the performance ofthe models is according to the significant analysis between the results of near infraredspectroscopy and standard methods. The results of SEC, RSQ, SEVC and1-VR evaluatethe prediction performance of PLS models, and we offer a new method “The analysis ofuncertain measurement” to verify the accuracy of ANN models.The PLS models of skim milk and whey powder exhibited good performances.The deviation of water content in skim milk reduced from0.07to0; SEC of watercontent in whey powder reduced from0.1055to0.0494, while SECV reduced from0.1117to0.0670, RSQ increased from0.4820to0.8584, and1-VR increased from0.4130to0.7414. We could conclude that after the correction, the models exhibited betterperformances. The ANN models of water content and fatty matter in buttermilk exhibitedgood performances. However, the deviation of protein content in buttermilk increased from-0.9to0.69, that means its ANN model exhibited bad performance.RSQ and1-VR of protein content and fatty matter in semi-skim and whole milkpowder approached to1; RSQ and1-VR of water content, protein content, fatty matter andash content in infant powder approached to1. We could conclude that their modelsexhibited good performances. However, RSQ and1-VR of pH in infant powder is0.8482and0.7851, its model exhibited not so good performance.RSQ and1-VR of dry matter in semi-condensed skim milk, condensed whey andcondensed milk infant approached to1; RSQ and1-VR of protein content insemi-condensed skim milk approached to1. We could conclude that their models exhibited good performances. However, RSQ and1-VR of protein content in condensed whey andcondensed milk infant is approached to0.9, its model exhibited not so good performances.The near-infrared spectrometer has been put into use and its real-time monitoring ofproduction is ideal.