Improving Crop Health Management: Federated Learning CNN for Spinach Leaf Disease Detection

This paper offers a unique method that combines convolutional neural networks (CNN) with federated learning to identify and classify illnesses in spinach leaves. 6,785 photos of spinach leaves, spanning five distinct disease classifications, make up our dataset. To protect data privacy and minimize the requirement for centralized data storage, we established federated learning among four clients. Our study suggests that the federated learning CNN technique is useful for recognizing and classifying illnesses in spinach leaves. With values ranging from 89.09% to 95.16% for precision, 87.47% to 94.64% for recall, 90.30% to 93.41% for F1-score, and 0.95 to 0.98 for accuracy, local performance measures for each client showed encouraging results. The global performance measures improved after aggregating model updates using federated averaging, with values for precision, recall, F1-score, and accuracy across the four clients ranging from 91.21% to 93.18%, 90.43% to 92.70%, and 0.96 to 0.97, respectively. Further examination of the macro, weighted, and micro averages yielded precision values of 92.51%, 92.57%, and 92.56%; recall values of 92.93%, 92.92%, and 92.91%; F1-score values of 90.81%, 90.89%, and 90.87%; and accuracy values of 92.66%, 92.71%, and 92.70%. These results show that our suggested federated learning CNN model can accurately identify and classify spinach leaf diseases while protecting data privacy and reducing the requirement for centralized data storage.