Federated Learning CNN for Smart Agriculture: A Modeling for Soybean Disease Detection

With the help of federated learning and CNNs, this research paper aims to identify and categorize soybean leaf diseases. This study uses a federated learning approach to classify disease severity into four classes using four clients. The benefit of this method is that it preserves data privacy during analytics, which is highly important in our digital age. The information we obtain from each local customer is used to conduct our analysis. Using federated averaging, we produce an aggregate global model. It is possible to assess the model's performance on a local and international level by calculating macro, micro, and weighted averages. As a result of the macro standard stressing class equality, clients yu₁ through yu₆ performed well with grades of 91. 42%, 91. 08%, 90. 92%, 93. 15%, and 93. 40. Based on the weighted average of those clients with class imbalance, 91. 43%, 91. 07%, 90. 92%, 93. 13%, 93. 41%, and 92. 86%, respectively. Finally, the micro average averaging all inputs to the model indicates its overall accuracy with the following percentages for each client: 91. 43%, 91. 06%, 90. 91%, and 93. 11%. Using federated averaging to transform local data into a global data model proved this approach feasible in a real-world scenario. Precision, recall, and F1-scores for the studied clients indicate high sensitivity and specificity of the model. A precise agriculture technique can be adapted to control soybean foliar disorders resulting from this study. As a part of our move towards more resilient and safe farm management approaches, this study helps us solve the urgent task of detecting and classifying diseases.