AI-Empowered IoT Data Collection via UAV in Rural Areas

Soil monitoring is essential for smart agriculture in remote rural areas with limited connectivity. It helps predict regional runoff, soil erosion, and weather impacts while promoting more efficient irrigation. Current artificial intelligence (AI) methods often struggle to adapt to heterogeneous environments and limited connectivity. This study presents a vertical federated architecture called multi-head split learning (MHSL), utilizing AI-powered devices onboard Unmanned Aerial Vehicles (UAVs) mission that is designed to increase awareness of in-situ soil moisture collected data to predict environmental trends for enhanced monitoring in rural areas. Our architecture connects the local convolutional neural network (CNN) head model of multiple worker UAVs to the long-short-term memory (LSTM) tail model of a central master UAV, creating a global model. This is made possible by adopting GPUs onboard and WiFi connectivity among UAVs. To validate our approach, we have used the real datasets of the TERENO-Wüstebach network. The numerical results show that our CNN-LSTM approach can predict the SSM data for the next days with sufficient accuracy measured in terms of mean square error (MSE). The good performance of CNN-LSTM has been supported by comparisons with other schemes in the literature.