Abstract:

Moving window partial least squares (MWPLS) is a method of regional optimization, which is most frequently
used for selecting spectral region including large information related to the components to be determined in the samples.
In this study, MWPLS was used to select the appropriate frequency range for setting up a partial least squares (PLS) model
for quantitative analysis of the total volatile basic nitrogen (TVB-N) in mutton. The near infrared reflectance (NIR) spectra
were processed by MWPLS and different spectral regions relevant to TVB-N content in meat were selected by changing the
window width of MWPLS. The evaluation and validation results showed that the optimal region for setting up a best PLS
model was the original spectrum between 1 325 nm and 1 484 nm. and the corresponding correlation coefficient of prediction
(Rp), standard error of prediction (SEP), rank and residual prediction deviation (RPD) were 0.856 84, 0.564 29 mg/100 g,
5 and 2.9, separately, suggesting that MWPLS is a valid method to select the spectral feature that reduces spectral data (from
800 reduce to 160) and enhances the prediction ability of the quantitative analysis model of TVB-N in mutton.

Key words: NIR spectroscopy, wavelength region selection, moving window partial least squares, TVB-N