Machine Learning-Based Multi-Frequency Water Content Prediction Using Microwave Resonant Cavity Sensor

To improve the accuracy of microwave resonant cavity sensors, which is hindered by low efficiency in the resonant frequency shift method due to sweep frequency detection, a multi-fixed frequency measurement method is explored. In this study, five operating frequencies (508, 633, 732, 795, and 1081 MHz) are determined based on the segmented multi-frequency concept. A threshold of −50dB is chosen for truncating the transmission coefficient (S21) amplitude, facilitating a simplify correlation between S21 amplitude and water content. Then, a multi-frequency full-range prediction model is established based on a fully connected neural network (FCNN). The model, structured with a 5-13-14-1 architecture, demonstrates test absolute errors of 99.52% within a ± 3% range and 93.72% within a ± 1 % range after extensive training. A high-precision, multi-fixed frequency method for measuring water content is provided.