Microwave permittivity-assisted artificial neural networks for determining moisture content of chopped alfalfa forage

Abstract

Moisture content of a commercially important forage biomass such as alfalfa (Madicago sativa), the specific alfalfa that we used in this research, is essential at various stages of production including harvesting, baling, storing, pelleting, and cubing. The moisture content of commodities such as green tea leaves also plays a key role in retaining natural taste and color during drying. The dielectric constants and the dielectric loss factors of chopped alfalfa were measured with moisture contents between 11.5% and 73%, wet basis (wb), and bulk densities ranging over 139 kg·m-3 to 716 kg· m-3. They were measured at 10 frequencies between 1 and 18 GHz at 20 °C with an open ended coaxial probe with 20 inputs for artificial neural networks. A three-layer neural network with 20 inputs, five hidden nodes, and one output for moisture content was built on an error back-propagation algorithm with momentum and adaptive learning techniques to predict the moisture content of alfalfa. It was independent of bulk density with a root-mean-square error of 1.09%. The prediction of moisture content of alfalfa independent of bulk density in 12 seconds exhibited the potential of this technique in measuring the moisture content of alfalfa and other medicinal and cash crops in batch and in production moisture measurements.