Tackling Agricultural Challenges of Red Globe Grapes Leaf Diseases: A Federated Learning CNN Approach

This study combines convolutional neural networks (CNN) with federated learning to detect and categorise six disease types in red globe grape leaves. The innovative approach uses distributed machine learning among four clients while protecting data privacy, which might have ramifications for other privacy-sensitive applications besides agricultural disease detection. Each client in the research had a local dataset of red globe grapes leaf diseases, and the study deployed a federated learning model across the four clients. The clients updated the global model collectively after individually training CNN models on their local data and sharing model updates with a central server while maintaining data privacy. The model's professional capacity to recognize and categorise illness states across clients with high reliability was shown by experimental outcomes assessed through Precision, Recall, F1-Score, Support, and Accuracy, supporting the effectiveness of federated learning. The model displayed a strong performance in terms of these average metrics, despite fluctuations in individual customers' results, which were caused mainly by possible class imbalances. Between 87.34% and 94.36% were the Macro-average values, 88.39% to 94.50% were the Weighted-average values, and 88.40% to 94.49% were the Micro-average values for all customers. These excellent results show the model's ability to balance Precision and Recall, adjust to class imbalances, and successfully detect true positives worldwide. This study offers a practical, scalable, and privacy-preserving approach for crop multi-class disease identification, making a substantial contribution to precision agriculture. A significant advancement in distributed machine learning has been made due to the encouraging findings highlighting the potential of federated learning with CNNs for more extensive applications.