(29) MLR analysis of rice by means of extended derivatives of NIR spectra
¡¡¡¡¡¡Proceedings of Japan-Thailand Joint Symposium on Nondestructive Evaluation
Technology, pp.103-110, 2004-5
In NIR analyses, first and second derivatives are sometimes used as important
pretreatments of spectra. There are, however, some drawbacks in derivatives
of spectra, such as deformation (shift or inversion) of peaks and noise
enhancement. To reduce or avoid these problems, two types of extended derivatives,
fractional derivative and fractional absolute derivative, are introduced
on the basis of Fourier transform approach of numerical differentiation.
The former extends the order of derivatives from an integer to any positive
real number and the latter has an additional property of suppressing the
shift and inversion of the peaks. Some fundamental properties of the extended
derivatives are elucidated using simple Gaussian peaks and NIR spectra
of rice. They are subsequently applied to MLR analysis of three major constituents,
water, amylase and protein, of rice to show their effect on the performance
of calibrations. It is shown that the multiple correlation coefficient
and the standard error of calibration depend on the order as well as the
type of derivatives, and that the best calibration result is mostly obtained
at a fractional order. It is also shown that, in general, selected wavelengths
in MLR are more stable in the FAD than in the FD, since the wavelength
of each absorption peak remains almost constant in the former and shift
with an increase in the latter.