Dielectric feature-driven dual prediction of moisture content and bulk density for rice quality assessment

Rapid and accurate determination of moisture content and bulk density in rice is essential for quality assessment, storage, and processing in the agri-food industry. In this study, a microwave-based sensing system was developed to simultaneously predict both moisture content and bulk density in rice cultivars using dielectric parameters as input features. Key dielectric features, including phase shift (phi), attenuation (A), dielectric constant (e'), loss factor (e"), and loss tangent (tans), were measured in the frequency range of 2-4 GHz. The nine machine learning regressors were implemented to develop predictive models, and their performance was comprehensively evaluated. The k-nearest neighbors and random forest models exhibited the highest predictive accuracy. The models achieved high accuracy in predicting moisture content (correlation coefficient (R) = 0.98, root mean square error (RMSE) = 0.015-0.024) and the most significant results for bulk density (R = 0.92, RMSE = 0.600-1.200). These findings provide potentially encouraging routes for integrating the method into larger digital-agriculture systems and state-of-the-art data storage infrastructure.