Using Dielectric Spectra And Spectral Technology To Predict The Ratios Of Cow's Milk In Goat's Milk

With the improved living standards and consumption levels,people pay more and more attentions to the quality.In recent years,goat's milk become people's first choice for milk,because of its high nutrition,easy absorption and difficult allergies.However,cheap cow's milk is usually added in goat's milk for greater profits.Traditional methods for detection of cow's milk are based on physical or chemical methods.These methods are time-consuming,hard to handle and high cost.RF/MF dielectric properties and Near-infrared(NIR)spectroscopy have been widely applied in food quality analysis in recent years,they are efficient,simple and can realize online detection.In this research,dielectric properties and near infrared spectrum of different content of cow's milk in goat's milk were acquired.Then the samples were divided into calibration and prediction set based on joint x-y distances method.Standard normal vitiate transformation was used to preprocess original spectra.The modeling methods,such as partial least squares(PLS),support vector regression(SVR)and extreme learning machine(ELM)were applied to establish models for predicting the cow's milk content in goat's milk based on characteristic variables selected by uninformative variable elimination(UVE),characteristic wavelengths selected by successive projections algorithm(SPA)and full spectra(FS),respectively.The conclusions were as follows:(1)For single sample,the frequency and the cow's milk ratio had great influence on goat's milk dielectric parameters.Within each test frequency,linear coefficients of correlation of ? ? and ? ? with the incorporation of cow's milk is greater than 0.97,especially when the frequency is less than 300 MHz,the linear coefficient of correlation is greater than 0.99.When the frequency became 2400 MHz,the linear coefficient of correlation of ? ? and cow's milk incorporation rate reaches the maximum,0.99.So under certain conditions,there was a good linear correlation between the rate of incorporation of cow's milk and goat's milk dielectric parameters.(2)In the process of using coaxial probe detection technology to predict the ratio of cow's milk in goat's milk,the number(17)of principal components extracted by PCA was only 4.0% of FS.The model was more simplified,and the PLS model built by SPA had the highest correlation coefficients in prediction set which was 0.944.(3)In the process of using NIR spectroscopy for the prediction of cow's milk in goat's milk,the number of principal components extracted by SPA was only 0.5% of FS.The model built based on SPA was more simplified,but the useful information of extraction was little and the performance of prediction models were poor.The number of characteristic wavelengths extracted by UVE was 192,which had too much input variables.From the point of view of modeling result,the best correlation coefficient of the ELM model for prediction set was 0.643,the correlation coefficients of FS-SVR and FS-PLS models were higher than 0.80 for both calibration and prediction set.Correct prediction rate of FS-SVR model was the highest,whose correlation cofficient reached 0.883 for prediction set.(4)The correlation coefficients of PLS model built based on dielectric spectroscopy and SVR-FS models which were higher than 88% for prediction set,were 0.944 and 0.883.The results of this research showed that the coaxial probe technology and near-infrared reflectance spectroscopy technique as an efficient,accuracy and non-invasive method could be used to quantitatively detect cow's milk in goat's milk.And the coaxial probe technology was the better method to do it.