WiMgrain: Multivariety Grain Moisture Content Detection Using Wi-Fi CSI Data

Moisture content detection is crucial in grain trade and storage processes. To address the high cost and invasive of traditional methods, inexpensive commercial Wi-Fi devices have been used for detecting the moisture content of grains. However, existing solutions are only designed for a single variety of grains. In this article, a new multivariety grain moisture detection method based on commodity Wi-Fi devices is proposed, termed WiMgrain, which can adaptively detect the moisture of different varieties of grains. The new method can improve the generalization ability and detection efficiency of the Wi-Fi signals in grain moisture detection. Specifically, we design a specialized Wi-Fi channel state information (CSI) data preprocessing method to address hardware challenges and environmental noise. Then, we propose a spatial equivalent permittivity (SEP) representation method based on multiantenna CSI to better represent grain moisture content information. Next, a new domain adversarial neural network (DANN) is proposed, termed ConTransDANN, which integrates the Conformer module to combat interference from morphological features of different varieties of grains. Finally, we evaluate the performance of the system through three different varieties of grains. The experimental results show that when each variety of grain is set with six different moisture contents, WiMgrain has an overall accuracy of 95.4% and a resolution of 1% for detecting grain moisture content in the line-of-sight (LOS) scenario.