Agricultural Innovation: Federated Learning and CNNs in Guava Leaf Disease Classification

To solve the pervasive problem of guava leaf diseases, this research article provides a ground-breaking study that uses Federated Learning (FL) in conjunction with Convolutional Neural Networks (CNN). The categorization of six common guava leaf diseases is the main emphasis of the research, which makes use of four different client datasets. Large-scale dataset management and use for illness diagnosis present considerable hurdles in the age of big data. Federated learning, a technique that enables model training across several decentralized clients while maintaining data privacy, is how our approach tackles these problems. The model was individually trained for each client using local data, and the model parameters were then combined to create a global model via federated averaging. For the global models from each client, the precision, recall, F1-Score, support, and accuracy varied from 91.73% to 93.67%, 91.68% to 93.68%, 91.70% to 93.66%, 875.17 to 1051.67, and 0.97 to 0.98, respectively. Additionally, the comparison study was done to comprehend the unique contributions made by each customer to the overall model. Macro-average, which calculates the average performance across all courses, varied from 91.70% to 93.66% for all customers. The Micro-average, the mean performance overall individual instances, fell between 92.06% and 93.99%, while the Weighted-average, which considers class imbalance, ranged between 92.07% and 94.01%.. The findings of this research provide significant contributions to science and have practical ramifications for the global and Indian agricultural sectors, resulting in improved disease control and higher crop yields.