Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: Robust Calibration for Determination Protein and Amylose in Rice Flour Using Nir Spectroscopy

Authors
item Sohn, Mi Ryeong - USDA-ARS, ATHENS, GA
item Barton Ii, Franklin

Submitted to: US Japan Nutritional Research Panel
Publication Type: Abstract Only
Publication Acceptance Date: December 1, 2002
Publication Date: December 1, 2002
Citation: SOHN, M., BARTON II, F.E. ROBUST CALIBRATION FOR DETERMINATION PROTEIN AND AMYLOSE IN RICE FLOUR USING NIR SPECTROSCOPY. THE 31ST UNITED STATES-JAPAN NUTRITIONAL RESEARCH PANEL. 2002. ABSTRACT P. 30-31.

Interpretive Summary: This is an abstract only. No intrepretative summary is required.

Technical Abstract: Robust calibration models were developed to determine protein and amylose content in rice flour using near infrared (NIR) reflectance spectroscopy. A total 220 rice samples from 1996 (90 samples) and 1999 (130 samples) were collected from US, Taiwan, Korea and Australia and used in calibration. Calibration samples were milled and then ground, and had a range of 4 to 14% for protein and 0 to 26% for amylose. The NIR reflectance spectral data of milled rice flour samples were scanned over the range of 1100 to 2498 nm at 2 nm intervals. All spectral data were preprocessed by "background correction" due to changes in the instrument due to repair. A partial least squares regression (PLSR) was processed using the various math treatments, such as multiplicative scatter correction (MSC), 1st or 2nd derivatives of Norris type and Savisky-Golay type, smoothing, mean normalization, and Marten's Uncertainly to effect model improvement. All PLS models were carried out using Unscrambler software and standard error of prediction (SEP) was estimated by full cross-validation. MSC processing improved the model, which reduced the number of factor and the SEP and increased the correlation coefficient (R2). However smoothing and normalization did not improved the model at all. Derivative processing improved the result with 2nd Savisky-Golay derivative. The 1st derivative is more effective than the 2nd derivative. There were no difference between the models using 1st Norris derivative and 1st Savisky-Golay derivative. A Marten's uncertainly test reduced the number of factor but did not improved the R2 and SEP. Marten's regression model using 1st (either Norris or Savisky-Golay type) derivative with MSC processing was the best for 2 constituents. For protein, the R2 between predicted and reference value was 0.980 and SEP was 0.246% with bias of -0.001% using 3 factors. The wavelengths with high regression coefficients were 2150 nm and 2066 nm due to the N-H stretching based on protein. For amylose, the R2 and SEP were 0.976 and 1.155% with bias of -0.003% using 8 factors. The wavelengths with high regression coefficients were 2272 nm, 2282 nm and 2328 nm, which is due to the C-H and O-H stretching based on starch and cellulose.

Last Modified: 7/22/2014
Footer Content Back to Top of Page