REGRESSION ANALYSIS OF MICROWAVE SPECTRA FOR TEMPERATURE-COMPENSATED AND DENSITY-INDEPENDENT DETERMINATION OF WHEAT MOISTURE CONTENT

Authors: Archibald, Douglas; TRABELSI, SAMIR - OICD; Kraszewski, Andrzej; Nelson, Stuart

Submitted to: Journal of Applied Spectroscopy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/10/1998
Publication Date: N/A
Citation: N/A

Interpretive Summary: The study provides mathematical evaluation of instrumental and data processing options for rapid, non-destructive measurement of wheat moisture content using microwaves at single and multiple frequencies. Data treatments are designed to simulate the extremes of temperature, bulk-density, and moisture level that might be encountered in on-line wheat tmoisture-content measurement of grain flowing through a pipe in a grain handling or storage facility. The work will be of interest to researchers, instrument manufacturers, and end-users, because it provides a clear description of the performance tradeoff for a comprehensive set of possible instrumental configurations. Moreover, the report identifies and characterizes data artifacts, and analyzes trends and sources of error to assess future instrumental and mathematical approaches for improving the method. There is also an estimate of how much improvement is possible. The report advocates a particular kind of moisture-content determination method that is accurate, simple to implement in hardware, and produces output that is simple to process because temperature compensation is included in the algorithm. Moreover, the method performance has been systematically verified over the range of conditions that might be found in the targeted on-line application.

Technical Abstract: Partial least-squares regression (PLSR) was used to generate wheat moisture content predictive models from eight-frequency microwave attenuation (A) and phase (P) spectra in the 10.36 to 18.0 GHz-range, as obtained by a free-space technique with a 10.4-cm thick sample. Spectra (n=379) were measured for a set of grain samples which had been treated to span the agriculturally practical ranges of moisture content (M) (10.6 to 19.2% g/g(wet)), temperature (K) (-1 to 42 C), and bulk density (D) (0.72 to 0.88 g/ml). The sample-property space formed by M, K and D was used to prune redundant samples, and a robust strategy was used to select subsets for calibration (n=279), cross-validation (n=40 segments), and testing (n=31). Twelve model types are reported and vary from A or P alone to the combination of A, P, K and D. To optimize each PLSR model, the raw spectral, K, and D data were preprocessed with variable ratios, mathematical transformations, and/or variable scaling. The lowest moistur prediction errors were for temperature- and density-corrected models with variables AKD or APKD; these produced root mean-square cross-validation and prediction errors (RMSECV and RMSEP) of 0.19 to 0.20% in moisture content units. The more practical unifrequency models, APK at 15.2 GHz, and AK at 18.0 GHz, yielded RMSECV of 0.21% and 0.35%, respectively. Addition of K to dielectric data always substantially reduced the model error. However, the multiplicative effect of density is well corrected by using the ratio A/P, or partly corrected by using the features in the A spectra. Data trends suggest that dual frequency PK models might benefit from a wider frequency range, and unifrequency AK models might be better at frequencies higher than 18.0 GHz.