Propelling Agritech Forward: Federated Learning and CNN for Sweet Potato Disease Diagnosis

In the current study, a unique method is suggested to address the problem of identifying illnesses in sweet potato leaves by combining federated learning with Convolutional Neural Networks (CNNs). This work advances the field by showcasing how federated learning, applied via local and global model training, may result in highly accurate and effective disease detection in sweet potato leaves. The approach was rigorously tested using four customers' local data sets, each including six different classes of sweet potato leaf diseases. The local models consistently achieved precision, recall, F1-Score, and accuracy scores in the mid-to-high 90s, successfully classifying all illness classes. For instance, the model for Client C earned a commendable F1-score of 97.02 for Class F, whereas the model for Client A displayed an equally good F1-score of 94.50 for Class. For instance, the global model assessment of the data from Client C produced an F1-Score of 95.55, showing the model's resilience in the precise identification of illness classifications. Macro, weighted, and micro averages were used to assess the global model's overall performance in the final step of the investigation. The model produced an F1-Score for the macro-average of 95.98 and an even more excellent F1-Score for the weighted and micro-averages of 96.02. In conclusion, the study effectively illustrates the potential of federated learning and CNNs in improving the detection and treatment of illnesses affecting sweet potato leaves. This strategy offers a very efficient, doable answer to farmers all over the globe, greatly enhancing the production and health of the world's agriculture.