Abstract:

The objective of this study was to develop a method to determine the fat content in yuba by near infrared (NIR)
spectroscopy combined with chemometrics. A total of 180 yuba samples collected at different occasions from different
production lines were tested by NIR spectroscopy. The diffuse reflectance spectra (4 000?10 000 cm-1) were collected using
an integrating sphere attachment. In order to eliminate the particle scattering and baseline drift, the NIR reflectance spectra
were preprocessed by 2nd order derivative with Savitzky-Golay. Backward interval partial least squares (BiPLS), synergy
interval partial least squares (SiPLS), searching combination moving window partial least squares (SCMWPLS) and genetic
algorithms partial least squares (GAPLS) were employed to extract informative variables and construct quantitative models
for the fat content in yuba. After comparison, the best model was obtained by GAPLS method with 143 data points. The
correlation coefficient (r) was 0.96 and the root mean square error of cross-validation (RMSECV) was 0.95 in calibration set,
and the r was 0.92 and the root mean square error of prediction (RMSEP) was 1.17 in prediction set. This work demonstrates
that variables extraction methods not only allow selection of the NIR informative variables for the fat content of yuba and
simplify the models, but also highlight the potential of NIR technique for assessing the quality of yuba on-line.

Key words: near infrared spectroscopy (NIRS), yuba, fat, variables extraction, quantitative detection