Skip to main content
ARS Home » Research » Publications at this Location » Publication #80339
Title: MOISTURE DETERMINATION WITH AN ARTIFICIAL NEURAL NETWORK FROM MICROWAVE MEASUREMENTS ON WHEAT

Author
item BARTLEY, PHILIP - UNIV GA DEPT AGRIC ENGR
item MCCLENDON, RONALD - UNIV GA DEPT AGRIC ENGR
item Nelson, Stuart
item TRABELSI, SAMIR - OICD

Submitted to: Instrumentation and Measurement Technology Conference Record
Publication Type: Proceedings
Publication Acceptance Date: 5/19/1997
Publication Date: N/A
Citation: N/A

Interpretive Summary: Moisture content of grain is the most important characteristic determining the length of time that grain can be safely held or stored without spoilage. Therefore, moisture content must be measured at harvest and whenever grain is sold. The selling price is affected by moisture content, not only because the costs for drying must be considered, but also because the water in the grain has less value than the grain dry matter. Since th dielectric properties (electrical characteristics) of grain are highly correlated with its moisture content, electronic grain moisture testers have been developed that provide quick moisture tests. However, samples from grain lots must be taken judiciously and then tested to obtain reliable moisture information on the whole grain lot. Recent research has shown that microwave measurements should be useful for monitoring moisture in flowing or moving grain. However, because of bulk density fluctuations in the moving grain, the data from measured microwave parameters must be processed in the best way possible to obtain reliable moisture data. This new research has shown that data taken on wheat can be processed by an artificial neural network, providing excellent results for the moisture content. The technique mimics the operation of the human brain, can be very rapid, and could be applied for on-line grain moisture monitoring, which would provide essential information for preserving grain quality and improving quality in processed grain products.

Technical Abstract: An artificial neural network (ANN) was used to determine the moisture content of hard red winter wheat. The ANN was trained to recognize moisture content in the range from 10.6% to 19.2% (wet basis)from transmission coefficient measurements on samples of wheat placed between two radiating elements. The measurements were made at 8 microwave frequencie3s (10 to 18 GHz) on wheat samples of varying bulk densities (0.72 to 0.88 g/cm**3) at 24 deg C. The trained network predicted moisture content (%) with a mean absolute error of 0.135.