Agricultural Intelligence: Federated Learning CNN's Models for Jute Leaf Disease Analysis

The research study performs an experimental analysis of a federated learning system that uses convolutional neural networks (CNNs) to detect and classify jute leaf diseases from many distributed data sources. We focus on five specific categories of jute leaf diseases, using local datasets from five individual clients to generate personalized models. The models are ensembled using federated averaging to create a robust global algorithm. Federated learning harnesses privacy and data security but enables the model to learn from various illness symptoms. The findings are promising, indicating that the federated learning model not only preserves but also improves the accuracy of illness categorization when moved from local to global data sets. The model's performance was assessed using precision, recall, F1-score, support, and accuracy parameters. The average accuracy, recall, and F1score increased from 78.86%, 78.7%, and 78.71% for mq_1 to 94.21%, 94.19%, and 94.2% for mq_5 for the five clients. The weighted means showed a similar improvement, with mq_1 starting at 79.38% and mq_5 reaching 94.22%. Micro averages had the same trend of increase, growing from 79.27% for mq_1 to 94.22% for mq_5. The statistical measures reveal increasing diagnostic precision of the model as the federated learning process advances. The research also examines the performance of federated averaging in aggregating local data sets for a global model. The global model accuracy achieves the best accuracy of local clients, up to 0.98. This shows how federated learning improves the global model by adding local data contributions. The research emphasizes the efficiency of federated learning and CNNs in solving the complex problem of jute leaf disease classification.