Cutting-edge Analysis of Sweet Potato Leaf Diseases: Leveraging Federated Learning and CNNs for Severity Evaluation

Abstract

This study proposes a unique method for categorizing and estimating the severity of illnesses that affect sweet potato leaves using federated learning and Convolutional Neural Networks (CNN). The work is especially noteworthy since it may apply to agricultural situations where data privacy is crucial. The federated learning model demonstrated great precision, recall, F1-Score, and accuracy across five severity levels of sweet potato leaf diseases for four customers in the local results analysis. For example, Client 1 had 98% accuracy across all severity levels, and such patterns were seen in other customers. By averaging the local client data, aggregated performance measures for examining worldwide outcomes were produced. With an average precision of 94.96%, a recall of 94.52%, an F1-Score of 94.61%, and an accuracy of 98% across all customers, these measures showed the model's strong performance. This outcome confirms that the federated learning strategy was successfully implemented to maintain good performance across several clients. Finally, average metrics such as Macro, Weighted, and Micro averages were used to examine the overall data. For example, the Weighted average recall was 94.91%, the micro-average accuracy was 93.91%, and the macro-average precision was 94.60%. These averages demonstrate the model's accuracy, precision, and suitability for use in real-world agricultural contexts, regardless of the distribution of classes. The study's findings support federated learning in the agriculture industry by providing a scalable, precise, and privacy-protecting system for categorizing the severity of sweet potato leaf diseases. It offers a potential direction for ongoing study and real-world applications in intelligent farming, enhancing crop health and productivity.

Topics & Concepts