determining the fat concentration of fresh raw cow milk using dielectric spectroscopy combined with chemometrics

Dielectric spectroscopy is usually used to obtain dielectric properties of materials. it not only has the advantages of high speed and little or no sample preparation but can also be used in in-line or in situ quality monitoring. to provide a quick and convenient method for milk fat concentration determination using dielectric spectroscopy, the dielectric spectra of 143 fresh raw cow milk samples were obtained from 20 to 4500 mhz. all samples were divided into calibration set and prediction set by using joint x-y distances sample set partitioning method. one hundred twenty-three, 9, and 40 dielectric variables were selected as effective variables (evs) by applying uninformative variable elimination, successive projection algorithm, and stability competitive adaptive reweighted sampling methods, respectively. the models of partial least squares regression (plsr), extreme learning machine (elm), and least squares-supporting vector machine (ls-svm) were built when the full dielectric spectra and selected evs were used as inputs. the results showed that the plsr model developed with full dielectric spectra was the best model on fat concentration determination with the root-mean-squares error of prediction set of 0.168%. the study tells that dielectric spectra have potential in exploring a milk fat concentration sensor used for in-line or in situ detection.