RFID-based Soil Moisture Sensor for Smart Agriculture: a Gaussian Mixture Model Approach

In this work, we present an RFID-based indirect soil moisture sensor based on the application of Machine Learning. More specifically, we suggest an unsupervised approach that does not require information about the real height and moisture levels. This approach can be of great interest in practical agricultural deployments, where the careful deployment of tags at specific depths within the soil is challenging. It allows an estimation of the posterior probability of moisture, based on the available Received Signal Strength Indicator (RSSI) and phase. The suggested method enables the RFID system to operate as a sensor by probabilistically quantifying measurement uncertainty, which is a key distinction from existing methodologies. In this paper, we focus on two differentiated moisture cases to show the validity of our approach. Future research will extend the proposed methodology to a wider set of moisture levels.