Measurement System Based on Microwave Pseudo Waveguide and LSTM Neural Network for Moisture Content of Rice Grains

In this paper, a measurement system based on microwave pseudo waveguide together with transceiver antennas is designed for moisture content of rice grains. The system is modeled and simulated in CST Microwave Studio, and the simulation results prove the feasibility of our scheme. On this basis, an experimental setup is made to verify the simulation results. The scattering parameters (S parameters) with multifrequency sweeping are used to characterize the interaction between the microwave electromagnetic field and rice grains. The collected S parameters are selected in frequency bands to retain the dominant frequency bands and combined with the Long Short-Term Memory (LSTM) neural network to establish a prediction model for moisture content of rice grains. The results predicted by the microwave pseudo waveguide method are compared with those obtained by the standard gravimetric method. The LSTM neural network model exhibits good performance (R² = 0.9915, RMSE = 0.0071, MAE = 0.0060) in predicting moisture content of rice grains (ranging from 0.85% to 29.39%). The proposed microwave method is nondestructive, fast, and accurate, and has the potential to enable online and portable measurement. The measurement system combining the microwave pseudo waveguide method with the prediction model based on the classical deep learning algorithm has a promising application in agriculture and food industry.