Cv. Greensill et Kb. Walsh, Optimization of instrumentation precision and wavelength resolution for the performance of NIR calibrations of sucrose in a water-cellulose matrix, APPL SPECTR, 54(3), 2000, pp. 426-430
Instrumentation for near-infrared (NIR) spectroscopic applications should b
e optimized for the intended application. The influence of signal precision
and wavelength resolution was considered for the application of the noninv
asive assessment [NIR 700-1050 nm, partial least-squares (PLS) calibration]
of the sugar content of fruit, with the use of a model system of sucrose s
olution on cellulose. The precision (as estimated at the maximum count of t
he spectrum) of an MMSI Zeiss spectrometer (Carl Zeiss Pty. Ltd.) was varie
d by altering the number of scans averaged per spectrum, as well as the sig
nal level. Wavelength resolution was varied between 8 and 20 nm (as estimat
ed of the 912 nm Ar peak) by adjustment of the entrance slit of a prototype
spectrometer employing a Hamamatsu S4874Q photodiode array as the detector
, constructed on an optical bench. PLS calibrations were developed from int
eractance spectra of 0-20% w/v sucrose solution soaked filter papers, and c
ompared on the basis of standard error of cross-validation (SECV) and coeff
icient of correlation-validation (R-v). The optimum measurement precision f
or calibration development was lower than expected at a coefficient of vari
ation (CV) of 0.02 [signal-to-noise ratio (SNR) 5000:1]. Calibration perfor
mance was poorer at a resolution of <8 nm full width at half-maximum (FWHM)
in the NIR region, but not significantly different at resolutions of betwe
en 8 and 20 nm. Further work is required to define the upper threshold of w
avelength resolution. We conclude that instrumentation for the application
of fruit sorting can have a relatively poor resolution, and can afford to o
perate at signal-to-noise levels considered low for a photodiode array dete
ctor.