Precision Phytopathology in Agriculture: A Federated Learning CNN Framework for Banana Leaf Disease Classification

The outbreak of several panicle and foliage diseases causes a major threat to agricultural production and sustainability. This article introduces the disease categorization method for three types of leaf disease that uses the CNN architecture through a federated learning framework to classify them into four dryness levels. In this, four stakeholders, who are data centers that contribute to this collaborative learning, demonstrate its effectiveness in tackling agriculture domains in a wide range of fields of agribusiness. Through profound experiments that we have illustrated and intensive studies of the resulting model, we found that this model is capable of coping with the local variations and transforming them into a highly global learning model. The results of our study shown remarkable accuracy in categorising diseases into four categories of severity: grade, class, grade, and critical. The proof that the macro critical point, micro critical point, and train approaches at the same time was given by the best performance of the model, which had the highest accuracy values. zs₁, zs₂, zs₃, and zs₄ customers mean percentages were 92. 74%, 95. 79%, 95. 9%, and 96. 22%, respectively, indicating uniform quality from all categories. The weighted averages, which also consider the distribution structure, broadly resulted in a similar pattern; the scores were 92. 74%, 95. 81%, 95. 88%, and 96. 2%. The aggregate individual classes of Micro were almost the same as the Weighted averages, making an indication that Micro's model was accurate in this way. The task shows that federated learning can be successfully applied to solving agriculture problems related to plant disease recognition and thus is closely tied to the broader discussion of sustainable farming strategies.