Edge AI with on-Device Learning Capabilities for Precision Agriculture

Efficient water management in agriculture plays a strategic role in ensuring the economic sustainability of production and, at the same time, the conservation of natural resources. The adoption of innovative monitoring and control techniques makes it possible to improve irrigation efficiency, reducing waste and ensuring that water is supplied according to the actual needs of the crops. The work reported in this paper addresses this problem by designing and prototyping an intelligent, low-power embedded system for on-device measurement and calibration of soil moisture in precision agriculture scenarios. The goal is to develop a platform composed of sensor nodes interconnected through an IoT network equipped with sensors that differ in terms of accuracy and cost, and capable of locally learning and improving measurement quality by integrating on-device learning techniques. The proposed system is based on the idea of performing on-device recalibration driven by reference values that are received intermittently or periodically. In particular, it leverages the TinyRBF algorithm, a recently proposed neural network architecture that can be executed on embedded nodes under tight energy constraints.