Federated Graph Attention Neural Networks (FeGAN) for Privacy-Preserving Crop Disease Detection
Muhammad Bello Kusharki, Bilkisu Larai Muhammad-Bello, Sanjay Misra, Muhammad Muktar Liman, Nachamada Vachaku Blamah · 2025
The integration of Artificial Intelligence (AI) and Internet of Things (IoT) technologies is transforming precision agriculture by enabling accurate, real-time detection of crop diseases while ensuring data sovereignty. The Federated AIoT framework, combined with Graph Attention Networks (GATs), processes decentralized sensor and image data locally, addressing privacy concerns and supporting sustainable practices. Using federated learning, the model aggregates local updates from each farm to build a global model that benefits from diverse environments without compromising individual data ownership. Evaluated on the PlantVillage dataset and synthetic data, the framework achieved 90% accuracy, 90% precision, 75% recall, an F1-score of 81%, and a Matthews Correlation Coefficient (MCC) of 67%, indicating strong early disease detection performance. This approach empowers farmers with data control and offers scalability across varied agricultural contexts, with future research focused on improving real-time scalability in resource-limited settings.