Spectral Data Analysis. The spectroscopic (NIR and mid-IR)
detectors receive light from the sample in the form of diffuse
reflectance after absorption, specular reflectance, and scattered light.
Only the diffuse reflectance contains chemical information, whereas
the latter two carry no useful information as there is no large variation
in particle size between starch and onion powder (based on the result
from Cilas 1090 particle size analyzer). The collected spectra might be
affected by systemic noise due to light scattering, variation in particle
size, and instrumental drift. Thus, to determine an accurate chemical
composition from spectroscopic measurements, the raw spectra must
be corrected at different levels by applying preprocessing methods.18
In this study, both FT-NIR and FT-IR spectral data were treated with
five preprocessing methods: three data normalization methods
(minimum, maximum, and range), standard normal variate transformation
(SNV), and multiple-scatter correlation (MSC). The
original FT-NIR spectra of all the samples and the original spectra
preprocessed by the SNV method are depicted in Figure 1, panels a
and b, respectively. The raw FT-IR spectra for each sample and SNVpreprocessed
spectra are shown in Figure 2, panels a and b,
respectively.