Severity Stratification of Brinjal Leaf Diseases: A Federated Learning CNNs Approach

A new method combining Federated Learning and Convolutional Neural Networks (CNNs) is explored in this study for the precise identification of Brinjal leaf diseases in agricultural settings. A total of 5448 photos were used to train and test the model, covering a wide range of illness severity from very mild (1-25%) to very severe (76-100%). The study of the data reveals interesting new perspectives. The results showed that nx_6 was the leading client, with a remarkable 94.77 per cent accuracy, 94.82 per cent recall and 94.79 per cent F1-score. Having achieved such a high degree of performance, the capacity to classify across all levels of severity is almost flawless. Client nx_2 had mediocre performance, on the other hand, with precision, recall, and F1-score all falling between 77% and 79% and accuracy at 89%. This indicates room for improvement in the model's sensitivity. The intermediate clients (nx_1, nx_3, nx_4, and nx_5) did very well, with precision and accuracy scores mostly in the mid-80s. This showed that they were good at finding illnesses and putting them into groups.Moreover, the examination of average performance measures - Macro Average, Weighted Average, and Micro Average - across the customers offered a comprehensive assessment of the model's effectiveness. The weighted average varied from 77.64% (nx_2) to 94.78% (nx_6), and the micro average ranged from 77.72% (nx_2) to 94.78% (nx_6), highlighting the varying but usually excellent performance across different data circumstances. In summary, this research not only exemplifies the state-of-the-art in plant disease identification utilizing Federated Learning and CNNs, but it also ushers in new possibilities in precision agriculture, particularly in the context of Brinjal farming.