Machine Learning-Enabled Liquid Recognition Based on Multiple Microwave Complementary Split-Ring Resonators

In this study, we propose a microwave sensor, which is composed of triple complementary split-ring resonators (CSRRs), for liquid recognition. Compared to a single CSRR, the sensor exhibits more resonant features, enabling high identification accuracy. The sensor aims to recognize liquids in common drink bottles without damaging the contents, offering significant potential for safety inspection. However, challenges arise due to error factors such as placement, dimensions of the bottles, and, in particular, uneven surfaces. To address these issues, we employ a feedforward neural network (FNN) to enhance accuracy. A total of 26 different liquid samples are tested, and for each sample, 160 measured <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">S</i><sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">21</sub> datasets are used for training the FNN parameters, while 40 datasets serve to validate the model. When the bottles are positioned upright, the trained model achieves a stable identification accuracy of 97% after 2000 epochs. To improve the performance which is limited by the air frustum groove between the bottle and the sensing region, we propose placing the bottles on the sensor with the cap facing downward. This configuration minimizes the air gap, as the relatively flat cap fits closely with the sensing area. With this adjustment, the trained model achieves 100% accuracy after 800 epochs, demonstrating improved accuracy and efficiency. The proposed identification method and sensor also have potential applications in chemical, biological, and medicinal liquid analysis.