Grape Leaf Disease Severity Analysis: Employing Federated Learning with CNN Techniques

Abstract

The Convolutional Neural Network (CNN) model used in this study uses federated learning to identify and categorize grape leaf diseases into four severity categories. The model may learn from several data sources while maintaining data privacy thanks to the federated learning technique. Four customers are used to assess the model’s performance, each contributing local model weights to the global model. The findings show consistent performance across clients and severity levels for detecting and classifying grape leaf diseases. With precision, recall, F1-Score, and accuracy values as high as 97.30%, 96.82%, 95.92%, and 0.98, respectively, Client 4 performed best for the local models. The performance of the global model for each client was consistent, with Client 4 attaining the most incredible precision (95.86%), Recall (95.56%), F1-Score (95.70%), and Accuracy (0.98) values. Three alternative averaging methods—Macro average, Weighted average, and Micro average—were used to assess the average performance of the global model. Results for the Macro average (Client 2) varied from 93.56% to 95.71%, those for the Weighted average (Client 2) from 93.54% to 95.74%, and those for the Micro average (Client 2) from 93.53% to 95.73%. The CNN model based on federated learning exhibits high consistency and accuracy in identifying and categorizing grape leaf illnesses across the four clients and severity levels. The findings show how well-federated education protects data privacy while allowing the model to learn from various data sources, thereby enhancing the model’s overall performance in detecting and classifying grape leaf disease.

Topics & Concepts