Submitted to: Cereal Chemistry
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/26/2001
Publication Date: N/A
Interpretive Summary: The protein and amylose contents of rice are important factors in the estimation of the quality of rice for various markets. However, the chemical analyses that have been traditionally used for the determination of these factors are environmentally unfriendly, time consuming and fraught with errors. The speed and precision of the chemical analyses have been improved by automating them but the accuracy has not been improved neither have the environmental problems been avoided. The use of spectral analyses that are more rapid, not subject to many of these errors and do not produce chemical waste have been sought. The use of near-infrared reflectance spectroscopy (NIRS) has overcome many of these problems but suffers from overlapping spectral bands and sensitivity to absorbed water. Near-infrared Fourier-transform (NIR-FT) Raman spectroscopy is another spectral technique that has the potential of providing the same information nas NIRS, but is relatively insensitive to water and has more definitive spectral bands. This study demonstrated that NIR-FT/Raman spectroscopy based models could be developed that were essentially equivalent in predictive ability to that possible with NIRS and have the additional advantages of insensitive to water, sharp bands and excellent stability. Further development of this method should make it a viable alternative for the assessment of rice quality.
Technical Abstract: The chemometric calibration of near-infrared Fourier transform Raman (NIR- FT/Raman) spectroscopy was investigated for the purpose of providing a rigorous spectroscopic technique by which to analyze rice flour for protein and apparent amylose content. Ninety rice samples from a 1996 collection of short, medium and long grain rices grown in 4 states of the United States, Taiwan, Korea, and Australia were investigated. Milled rice flour samples were scanned in rotating cups with a 1064 nm NIR laser at 500mW of power. Raman scatter was collected and focused on a Ge (LN2) detector. Data over the Raman shift range of 160-3600 cm**-1 was utilized in the study. The spectral data was preprocessed using derivatives or baseline correction and normalization. Nearly equivalent results were obtained using all of the preprocessing methods using partial least squares models. However, models using baseline correction and normalization showed slightly ybetter performance and thus were deemed better for use. The best model fo protein (n=86) was obtained using 7 factors and gave r**2=0.988 with SECV=0.157% and with a bias=0.0003%. The best model for apparent amylose (n=86) was obtained using 14 factors and gave r**2=0.983 with SECV=1.05% and a bias=0.005%.