Sustainable Coffee Production: A Federated Learning Framework with CNN for Disease Detection and Classification

Using federated learning with Convolutional Neural Networks (CNN), this research study provides a novel technique for identifying and categorizing coffee leaf illnesses. The research focuses on five categories of coffee leaf diseases and includes information from five clients, combining local knowledge with global comprehension. The research uses three averaging techniques-Macro, Micro, and Weighted averages-in the Result Analysis section, each offering significant insights into the illness categorization. Precision was 83.53%, recall was 87.04%, F1-score was 91.23%, Support was 90.46%, and accuracy was 92.79%, according to the study of the macro Average. A precision of 85.73%, recall of 88.10%, F1-score of 91.74%, Support of 90.62%, and accuracy of 92.79% were obtained from the weighted Average. The Micro average's final statistics were accuracy at 92.78%, recall at 88.12%, F1-score at 91.73%, and Support at 90.62%. Additionally, federated averaging, a technique that harmonizes local data and transforms it into useful global insights, was used in the research's part on converting local data to global data. The values for the five clients were Cta-1: 83.69% precision, 83.41 % recall, 83.49% F1-score, 430.40 support, and 0.94 accuracy; Cta-2: 87.18%, 86.90%, 87.03%, 518.40, and 0.95; Cta-3: 91.30%, 91.16%, 91.21 %, 638.60, and 0.97; Cta-4: 90.68%, 90.25%, 90.44%, 757.00, and 0.96; Cta-5: 92.85%, 92.74%, 92.78%, 949.80, and 0.97. This research pioneers a unique approach to disease identification in coffee leaves by the synergistic use of various approaches, possibly revolutionizing agricultural practices and resulting in more sustainable, data-driven choices in the coffee sector.