Secure and scalable federated learning for citrus disease detection with CNN-ANN and data privacy for agriculture

The increasing complexity of citrus diseases poses a significant threat to global agriculture, leading to economic loss and food insecurity. Traditional centralized disease identification methods suffer from data privacy concerns, computational inefficiencies, and poor scalability. Although deep learning models offer high accuracy, they require vast datasets that cannot be centrally collected owing to privacy restrictions. To address these challenges, a new federated learning framework is proposed that leverages a hybrid CNN-ANN model for citrus disease detection. Federated learning enables multiple clients to collaboratively train a global model without sharing sensitive data and preserving privacy, while improving performance. The hybrid model combines Convolutional Neural Network (CNN) for spatial pattern extraction with Artificial Neural Network (ANN) for advanced feature learning to enhance classification accuracy. The proposed system achieved a global test accuracy of 99.44% with a loss of 0.61, thereby demonstrating its effectiveness in real-world applications. In addition, it is efficient in distributed networks and scalable for large agricultural deployments. Security is further reinforced by decentralized data processing and encryption techniques. This study presents a privacy-preserving, scalable, and highly accurate solution for real-time citrus disease monitoring.