研究目的
To assess the feasibility of Fourier-transform mid-infrared spectroscopy to determine milk coagulation and acidity traits of sheep bulk milk and to classify milk samples according to their renneting capacity.
研究成果
FT-MIR spectroscopy shows potential for screening sheep bulk milk based on coagulation ability, with better performance for acidity traits than coagulation properties. It can help identify milk suitable for cheese production, but accuracy is insufficient for replacing standard methods, necessitating further research.
研究不足
The prediction models had moderate to low accuracy, with ratio of prediction to deviation below 2.5 for milk coagulation properties, indicating they cannot replace reference laboratory methods. The study is limited to bulk milk from specific breeds and regions, and further research is needed for industrial application.
1：Experimental Design and Method Selection:
The study used Fourier-transform mid-infrared (FT-MIR) spectroscopy to predict milk coagulation properties (rennet coagulation time, curd firming time, curd firmness) and acidity traits (pH, titratable acidity) in sheep bulk milk. Partial least squares regression analysis was employed for prediction models, and partial least squares discriminant analysis for classification based on renneting capacity.
2：Sample Selection and Data Sources:
A total of 465 bulk milk samples were collected from 140 single-breed flocks of Comisana and Sarda sheep breeds in Central Italy during a routine milk quality payment program between January and April
3：List of Experimental Equipment and Materials:
20 Equipment included a MilkoScan FT6000 for milk composition and FT-MIR spectra, a Fossomatic FC for somatic cell count, a potentiometric pH meter (Mettler Delta 345), a Crison Compact D meter for titratable acidity, and a formagraph for milk coagulation properties. Materials included calf rennet (Caglificio Clerici Spa-Sacco Srl) and NaOH solution.
4：Experimental Procedures and Operational Workflow:
Milk samples were refrigerated and transported for analysis. FT-MIR spectra were obtained, and reference values for traits were measured using standard methods. Spectral regions with low signal-to-noise ratio were discarded. Data were split into calibration (75%) and validation (25%) sets for model building and testing.
5：Data Analysis Methods:
Partial least squares regression and discriminant analysis were performed using R software (pls and DiscriMiner packages). Goodness-of-fit statistics included coefficient of determination, root mean squared error, ratio of prediction to deviation, bias, and accuracy measures for classification.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
