Predicting moisture content and husked rice yield using electrical parameters and machine learning

Rice has been one of the most important crops in the world. Among them, the rapid on-site testing of husked rice yield and moisture content can greatly facilitate to acquisition and grade in the paddy field, thereby reducing the losses during transportation, storage, and processing. This study aims to identify the optimal model for the prediction of the moisture content and husked rice yield of paddy. An LCR meter was used to test six electrical parameters (dielectric constant, capacitance, dielectric loss, dielectric loss tangent, conductivity, and electrical conductivity) of paddy samples with the moisture content ranging from 9.94% to 23.51% at frequencies between 1 and 8 MHz. Subsequently, a rice huller was used to dehusk the paddy samples, in order to determine the husked rice yield. The datasets were then obtained from the electrical parameters, moisture content, and husked rice yield. The electrical parameters at 1 MHz were used to train the regression models, in order to predict the moisture content and husked rice yield of paddy. All electrical parameters at 1-8 MHz were used to train four machine learning models: neural networks, decision trees, support vector machines, and random forests. The models were evaluated for correlation and accuracy using determination coefficients (R²) and root mean square errors (RMSE). Results indicated that the dielectric constant and capacitance initially decreased and then increased with increasing frequency. While the dielectric loss decreased consistently. The dielectric loss tangent initially decreased, then increased, and finally decreased again after 6 MHz. Conductivity and electrical conductivity increased consistently. All six electrical parameters increased with the rising moisture content. The regression model was also established using electrical parameters and moisture content. Therefore, the moisture content also increased under the condition of 1 MHz, as the six electrical parameters increased. The determination coefficients of moisture content with the six electrical parameters were ranked in descending order of conductivity, capacitance, dielectric constant, electrical conductivity, dielectric loss tangent, and dielectric loss. Among them, the fitting model with the highest determination coefficient for the conductivity and moisture content was 0.960, with an RMSE of 0.89%. In the fitting model of moisture content and husked rice yield, the determination coefficient was 0.944, with an RMSE of 3.05%. The husked rice yield of paddy increased first and then decreased with rising moisture content. Moreover, the husked rice yield reached the maximum of 69.31% at a moisture content of 13.34%. According to the fitting model of electrical parameters and husked rice yield, the determination coefficients of the six electrical parameters with the husked rice yield were ranked in the descending order of the conductivity, dielectric constant, capacitance, electrical conductivity, dielectric loss tangent, and dielectric loss. Notably, the best fitting was achieved in the conductivity and husked rice yield, with a determination coefficient of 0.929 and an RMSE of 3.08%. Additionally, the neural network achieved the highest performance in predicting the paddy moisture content and husked rice yield among the machine learning models, with the determination coefficients of 0.987 and 0.935, and RMSEs of 0.87% and 1.87%, respectively, indicating the best regression. As such, the neural network model can effectively establish the relationship between electrical parameters and rice quality, thus accurately predicting the moisture content and husked rice yield of paddy. Machine learning was highlighted for the non-destructive testing of agricultural products, in terms of speed and efficiency. The rapid and accurate assessment was also performed on the key quality indicators, such as moisture content and husked rice yield. Meanwhile, this convenient and efficient approach can greatly enhance the efficiency of paddy management for cost savings in the supply chain. The valuable theoretical insights can also provide for the non-destructive quality assessment of the paddy, particularly in the broader application of advanced analytical techniques in agricultural science and industry.