Federated Learning CNN's in Agriculture: Unveiling Disease Severity Analysis of Java Plum Leaf's

The unique approach introduced in this research categorizes Java Plum Leaf illnesses into five severity categories using an FL_CNN. The study tackles the primary difficulty of accurate diagnosis and early response to these plant diseases by leveraging four local clients (Cl_X, Cl_Y, Cl_Z, and Cl_W). Operating across local clients, the federated learning network showed a surprising performance. For instance, in Cl_X, F1-Scores were close to 90% across all severity classes, and the model's performance varied from an accuracy of 95% to 97%. Class E highlights the model's ability to accurately identify genuine positive cases in Cl_X, which had the highest recall value of 93.50%. Like customer Cl_Y, who showed improved outcomes, Class A displayed an enhanced 94.92%, 92.57% and 98%. The values for all classes in Cl_Y varied from 90.33% to 94.57%., supporting the model's effectiveness even further. Results from the client datasets Cl_Z and Cl_W were similar. Class A achieved excellent precision values (94.34% in Cl_Z and 94.92% in Cl_W) in both datasets, resulting in respectable F1-Scores of 91.75% and 92.57%, respectively and an impressive accuracy of 98%. Additionally, the global performance of the federated learning CNN supported the regional analyses. For example, Cl_X had a vital accuracy of 96% and an F1 score of 90.99%. The F1-Scores for Cl_Y, Cl_Z, and Cl_W ranged from 92.33% to 92.73%., and all three had constant accuracy values of 97%. In its conclusion, this research emphasizes the reliability of the federated learning CNN in correctly identifying Java Plum Leaf illnesses across a range of severity levels. Its outstanding performance on a local and worldwide scale demonstrates its ability to impact plant pathology substantially.