At the IEEE International Conference on Artificial Intelligence of Things (IEEE AIoT 2025), held in Osaka, Japan, from December 3 to 5, 2025, the paper "Edge-Based Predictive Data Reduction for Smart Agriculture: A Lightweight Approach to Efficient IoT Communication" was presented.
The paper was authored by Dora Kreković, Mario Kušek, and Ivana Podnar Žarko from the Faculty of Electrical Engineering and Computing (FER), University of Zagreb, in collaboration with Danh Le Phuc from the Technical University of Berlin, reflecting a strong international research partnership between the institutions.
Accepted following an international peer-review process, the paper was presented on-site by Dora Kreković. The research addresses key challenges in smart agriculture by introducing a lightweight predictive edge AI approach that performs local inference and transmits data to the cloud only when meaningful deviations occur. Experimental results demonstrate communication reductions exceeding 92%, while maintaining high prediction accuracy. The study evaluates multiple deployment scenarios, including in-situ learning, cross-site model transfer, and satellite-to-ground inference, highlighting how satellite data can support deployments in locations without historical sensor measurements. Overall, the work underscores the potential of edge intelligence to enable scalable, efficient, and resilient smart agriculture systems, particularly in environments with limited connectivity.
