Advanced Mango Leaf Disease Detection and Severity Analysis with Federated Learning and CNN

Abstract

Worldwide mango production is threatened by mango leaf diseases, resulting in considerable financial losses. Effective illness management and reduction depend on early identification and severity level categorization. This research provides a brand-new FL-CNN model for categorizing mango leaf diseases into four severity categories. The FL-CNN model allows training to be distributed across several clients without compromising data privacy. The suggested model consistently outperforms all clients in identifying the four severity levels of mango leaf diseases. The model's efficiency across customers with various data distributions is shown using macro, weighted, and micro-averaging techniques. Client 4 had the greatest macro average (96.11), weighted average (96.09), and micro average (96.08) values, while Client 1 had the lowest macro average (93.92), weighted average (93.96), and micro average (93.96) values. These findings show that the FL-CNN model successfully recognizes and categorizes various degrees of mango leaf disease severity, making it an invaluable tool for practical agricultural applications. Utilizing federated learning provides data privacy while offering a scalable and safe method for managing plant diseases. Additionally, federated learning preserves data privacy and lessens the need for data centralization, offering a solid foundation for cooperation between many stakeholders. Future studies should focus on enhancing the model's effectiveness and accuracy and investigating whether it can predict additional plant diseases and agricultural situations.