Skip to main content
ARS Home » Research » Publications at this Location » Publication #90897
Title: MULTIVARIATE ANALYSIS OF MICROWAVE SPECTRA FOR SIMULATION OF DEDICATED INSTRUMENTS FOR TEMPERATURE-COMPENSATED AND DENSITY-INDEPENDENT ONLINE DETERMINATION OF WHEAT MOISTURE CONTENT

Author
item Archibald, Douglas
item TRABELSI, SAMIR
item KRASZEWSKI, ANDRZEJ
item NELSON, STUART

Submitted to: Pittsburg Conference on Analytical Chemistry and Applied Spectroscopy
Publication Type: Abstract Only
Publication Acceptance Date: 3/4/1998
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: This study used multivariate regression techniques to systematically evaluate various online instrumental configurations and data preprocessing methods using measurements of wheat grain that were made in a laboratory. Partial least-squares regression (PLSR) was used to generate moisture content predictive models from eight-frequency microwave attenuation (A) and phase (P) spectra in the range 10.36 - 18.0 GHz, as obtained by a free-space technique using an 11 cm thick sample cell. Spectra (n=379) were for a set of grain which had been treated to span the agriculturally practical ranges of moisture content (M) (10.6 - 19.2 %, g/gwet), temperature (K) (-1 - 42 C), and bulk density (D) (0.72 - 0.88 g/ml). The model types reported vary from A or P alone to the combination of A, P, K and D. Variable ratios, mathematical transformations, and variable scaling were applied to the raw spectral, K and D data in order to optimize each PLSR model. The lowest moisture error was obtained with a temperature-and density-corrected model (using variables AKD), and produced root mean-squared cross-validation and prediction errors (RMSECV and RMSEP) of 0.19 - 0.20 % moisture content. Compensation with an independent measure of temperature (K) is worth the additional instrumental complexity, but density fluctuation interference is much smaller than K interference and is well corrected using the ratio A/P. Moreover, full microwave spectra are not required for good predictions; the unifrequency models APK at 15.2 GHz, and AK at 18.0 GHz yielded RMSECV of 0.21 and 0.35 %, respectively. Data trends suggest that dual frequency PK models might be improved with a wider frequency range, and unifrequency AK models might be better at frequencies higher than 18.0 GHz.