Determination On Routine Nutrients And Establishing NIR Rapid Prediction Model For Bactrian Camel Milk

Bactrian camel milk is very nutritious,and with significantly higher fat,lactose and total dry matter than cow's milk,thus receiving more and more attention and industrializing and commercializing step by step in China,especially after its hypoglycemic effect and diabetes auxiliary treatment activity have been approved.Therefor its quality and safety inspection and supervision are imperative.To-date,routine nutrients of Bactrian camel milk still lack near-infrared(NIR)rapid prediction model.A total of 153 Bactrian camel milk samples were collected from Wuhai,Alxa Left Banner and Hulun Buir of Inner Mongolia,and 28 Holstein cow milk and 40 buffalo milk from Nanning Guangxi were collected to establishment and evaluation on a general prediction model.Fat,lactose,protein,water and ash contents were analyzed according to national standard method,and each milk sample were scanned by a desktop Fourier transform NIR spectrometer and a potable micro NIR spectrometer to obtain diffuse transmission NIR spectrum.Chemometrics software Unscrambler10.5 was applied to pretreat the NIR spectrum and PLS module used to calibrates each nutrient value and the NIR spectrum to established the milk five routine nutrients NIR prediction model.Model were optimized and choose by Internal validation parameters R2 and RMSECV,and robustness and accuracy of prediction were evaluated by external validation parameters such as RMSEP,RPD and RSD.General prediction model for five nutrients of cow,buffalo and bactrian camel milk,the optimized R2 were all above than 0.9,however the RPD were all below 2.5,this mean the general model applicable but no ideal,and this suggest that a specialized model needed to bactrian camel milk.The optimum parameters of desktop NIR model for bactrian camel milk were as follow:R2,RMSECV,RMSEP,RPD and RSD to fat prediction 0.9646,0.2036,0.1759,4.3206 and 3.37%respectively;to lactose 0.9223,0.1403,0.1056,3.3568 and 2.09%;to protein 0.9999,0.2010,0.0955,5.0387 and 2.66%;to moisture 0.9952,0.1510,0.1121,12.8670 and 0.21%;and to ash 0.9337,0.0202,0.0075,3.1642 and 0.85%.R2 were all above 0.92,and RPD were all higher than 3,this means the model set is robust and accurate.The correlative NIR spectrum pretreat methods for the optimum model of fat,lactose,protein,moisture and ash were SNV,MSC,second derivative,SNV and MSC+SNV respectively.When it comes to the potable micro NIR spectrometer,for the optimum prediction model,R2 were all above 0.9,RPD were all higher than 2.4,but below 3,it means its prediction accuracy is not as good as the desktop NIR spectrometer,but applicable in the field.