RF-TMS-CNNam: Non-Destructive Passive Liquid Concentration Detection Using RFID Systems

It is essential to accurately detect liquid concentration in the production processes. Radio frequency identification (RFID) is commonly used for item identification and detection due to its sensing capabilities. This article introduces the radio frequency-two-stage weighted multipath suppression convolutional neural network based on a hybrid attention mechanism (RF-TMS-CNNam) system, a novel approach that combines RFID technology and deep learning models to accurately determine the concentration levels of various liquids. The propagation of radio frequency (RF) signals through the target liquid is influenced by concentration levels, resulting in distinct changes in both phase and received signal strength indication (RSSI). These changes in phase and RSSI are utilized as features to detect the target. Subsequently, the features are extracted and fed into the convolutional neural network (CNN) to train a model for classifying different liquid concentrations. The problem of multipath interference is common in RFID sensing, distorting the sensing phase and RSSI and affecting the accuracy of target sensing. This article proposes a two-stage weighted multipath suppression (TMS) method to tackle this issue. In the first stage, the process identifies the channel with the least multipath interference. In the second stage, it assigns different weights to signals detected by various channels and tags. This approach significantly enhances the power of the direct path signal while reducing the power of multipath interference to an average value due to its randomness. After suppressing the multipath interference, “clean” features are extracted to train the liquid concentration detection model. Experiments were conducted to evaluate the performance of the proposed system. The experimental results show that the RF-TMS-CNNam system achieves an average accuracy of 97.8% and 96.8% in detecting the concentration levels of both white wine and wine. The proposed method also enables non-destructive object detection and outperforms other object detection techniques in terms of accuracy.