Journal
FOOD CONTROL
Volume 82, Issue -, Pages 57-65Publisher
ELSEVIER SCI LTD
DOI: 10.1016/j.foodcont.2017.06.015
Keywords
Multivariate calibration; Wavelength selection; Spectral preprocessing; Feed and food; Chemometrics
Categories
Funding
- Ministry of Agriculture Strategic Research Chair (PY) Programs
- Saskatchewan Agriculture Strategic Research Program Fund
- Agricultural Development Fund (ADF)
- SaskMilk
- Saskatchewan Forage Network (SNK)
- Western Grains Research Foundation
- SaskPulse Growers
- Natural Sciences and Engineering Research Council of Canada (NSERC)
Ask authors/readers for more resources
Growing trends towards the development of rapid, simple, and cost-effective methodologies for grain quality monitoring have been observed for a long time. Infrared spectroscopy associated with chemometrics has proven to be a promising alternative to traditional wet chemistry methods. So far comparative research that focuses on the relation between spectral preprocessing, wavelength selection, and model performance is rare. This paper aims to explore the effect of spectral preprocessing and wavelengths selection on the performance of partial least square (PLS) regression models constructed with near-infrared (NIR) and mid-infrared (ATR-FT/MIR) spectroscopy. Eighty wheat samples were randomly selected in Western Canada. NIR (680-2500 nm) and MIR (4000-700 cm(-1)) spectra were collected and calibrated with reference crude protein (CP) and moisture content using PLS regression technique based on different spectral preprocessing methods and selected important wavelengths. In a general way, the models generated with NIR spectra showed superior performance than that developed with MIR spectra. For CP, the best NIR model was developed with the 1400-2500 nm spectral range of standard normal variate (SNV) and first derivative (FD) pretreated spectra, which showed an excellent prediction performance (R-2 = 0.97); the best MIR model was established using the 1750-1100 cm(-1) region of SNV corrected spectra, which gave a good predictive ability (R-2 = 0.90). Regarding to the moisture, the best model obtained by NIR technique (based on the 1100-2500 nm region of raw spectra) showed a good prediction performance (R-2 = 0.86), while the best model generated by MIR technique (using the full wavenumber range of FD-SNV preprocessed spectra) only gave approximate quantitative prediction (R-2 = 0.72). Both spectral preprocessing and wavelength selection could affect the performance of PLS models for predicting CP and moisture content in wheat. More effort is needed to improve the performance of ATR-FT/MIR spectroscopy for quantitative analysis based on appropriate chemometrics tools. (C) 2017 Elsevier Ltd. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available