RF-LqRNN:RFID-Based Concentration Detection of Alcohol Solutions and Glucose Solutions

Solution concentration detection plays an important role in many real-life applications. For example, alcohol concentration detection can prevent cases where low-concentration liquor red is used to replace high-concentration liquor, while glucose concentration detection can be used to prevent diabetes. In this article, the radio frequency-liquid recurrent neural network (RF-LqRNN) system is proposed, which can classify solution concentrations using an inexpensive commercial off-the-shelf radio frequency (RF) identification device. The proposed system uses the feature that the phase and received signal strength indicator (RSSI) of RF signals across solutions of different concentrations are different. Measurements of the two features were obtained at different concentrations and input into a gated recurrent unit (GRU)-recurrent neural network (RNN) to train a solution concentration classification model. The experimental results show that 98.5% classification accuracy can be achieved for ten alcohol solution concentrations, and 97.3% classification accuracy can be achieved for ten glucose solution concentrations. The effects of ceramic, glass, paper, and plastic containers on solution concentration classification are also investigated, and it is determined that the solution concentration classification accuracy is the lowest in ceramic containers and the highest in plastic containers. Even at the lowest accuracies achieved in ceramic containers, the proposed method achieves more than 94% in the classification of the concentration of alcohol and glucose solutions.