Boosting Microwave Hydration Sensors Performance with Machine Learning Techniques

This study introduces innovative non-invasive hydration microwave sensors combined with machine learning (ML) algorithms for monitoring purposes. Specifically, a miniaturized complementary-split ring resonator (CSRR) operating in the 2÷ 3 GHz band for assessing human skin hydration through near-field interaction, and a separate patch antenna operating at 2.45 GHz for tracking the drying process of greenwood. The design process involves full-wave simulations to evaluate the resonators and patch antenna capability to effectively penetrate skin layers and wood structures. After fabrication on cost-effective substrates, extensive testing measurements were conducted on different mediums. Human volunteers' proximal wrist areas are monitored over six days with multiple daily measurements under various dietary conditions. Concurrently, the greenwood sample is assessed over twenty days in a controlled climate chamber. The spectral data obtained from the resonator and the patch antenna are analyzed using advanced multivariate data analysis (MVDA). The results confirm the method's effectiveness in accurately categorizing hydration levels and emphasize its potential for practical hydration monitoring applications due to its cost-effectiveness and operational efficiency.