The Role of Federated Learning and CNNs in the Detection and Classification of Zucchini Leaf Diseases: A Severity-Based Analysis

This study investigates using FL_CNN to categorize zucchini leaves into five severity levels. In this investigation, five customers representing various severity levels were used. The global model was iteratively trained and updated using CNNs deployed on client devices as part of The experimental technique. The analysis results demonstrated the model’s effectiveness and robustness, managing the impacts of increasing severity levels on complexity while retaining good accuracy to values across all clients. With macro, weighted, and micro averages for 86.04% to 94.29%, 86.69% to 94.42%, and 86.33% to 94.35%, respectively, while maintaining an overall high accuracy of 95-98%, the local data client’s performance notably illustrated the value of the Federated Learning model. The investigation also looks at global results, where the performance of the worldwide model, which integrates all local models, is evaluated. The global model performed well, with macro, weighted, and micro averages between 86.35% and 94.40%, despite the inherent complexity associated with varying sickness severities. The model’s ability to provide a high level of precision, recall, and overall accuracy across all severity levels was highlighted by this result. Subsequent analysis confirmed the global model’s ability to effectively learn from various data sources, making it a crucial tool in precision agriculture.