The Regression Analysis And Pattern Recognition Of Milk Based On Infrared Spectroscopy

Evaluation of milk quality is a main issue that the dairy industry is focusing on. In orderto ensure milk quality, the country established a series of standards. However, the traditionaldetection methods are time consuming, reagent wasting and environmental unfriendly. In thisstudy, ATR-FTIR technique combined with chemometrics method was used to establish therapid qualitative identification and quantitative analysis models of milk, which could providean approach for realizing the geographical origin traceability, online analytics and authenticitydetection of dairy products.The main research contents were as followed: Different origins of commercial milk wereconsidered, ATR-FTIR technique was used to collect the spectra of samples and the samplingparameters were determined as followed: scan number32, resolution4cm^-1. Differentspectral data preprocessing methods and spectral regions were considered. The optimizedSIMCA recognition model according to the origin of milk was established in the spectralregion of2990-2850cm^-1,1800-1700cm^-1and1600-900cm^-1, through the automatic baselinecorrection and Savitzky-Goaly polynomials of9points smoothing. Under the significancelevel of α=5%, the identification rate of model for samples were above80%, and rejectionrate were above93%. This indicated that ATR-FTIR combined with SIMCA would be a validmethod for origin traceability of milk.Different categories of commercial milk were considered, the optimal models wereestablished after spectral preprocessing and region selecting. The protein, fat and lactose PLSmodels were established separately in the spectral region of2990-2850cm^-1,1800-1700cm^-1and1600-900cm^-1, through the automatic baseline correction and1st derivation. The R2between the predicted values and actual values of protein, fat and lactose models were0.948,0.961and0.989, respectively. This indicated that ATR-FTIR combined with PLS would be avalid method for rapid analysis of main contents in milk.Different concentrations of adulterated milk were considered. The SIMCA model wasestablished in the spectral region of2990-2850cm^-1,1700-1655cm^-1and1600-900cm^-1,through the automatic baseline correction and Savitzky-Goaly polynomials of9pointssmoothing. Under the significance level of α=5%, the models’ detection limit of urea andglucose were0.010%（w/v）, the recognition rate and rejection rate were above87%.Furthermore, the urea and glucose PLS models were established separately in the spectralregion of1700-1655cm^-1+1600-900cm^-1and2990-2850cm^-1+1500-900cm^-1. The R²between the predicted values and actual values of urea and glucose models were0.945and0.942, respectively. The results indicated that ATR-FTIR combined with SIMCA and PLSmethods could be a valuable analytical tool for authenticity of milk.