Spinach Leaf Disease Detection and Severity Analysis: Breaking New Ground with Federated Learning and CNN

Abstract

Early diagnosis of illnesses affecting spinach leaves is essential for maintaining agricultural output and food security. The accuracy and scalability of conventional disease detection techniques, such as human inspection and remote sensing, are constrained. Convolutional neural networks (CNNs) and federated learning are used in this study to categorise spinach leaf illnesses into four severity categories. Four customers participated in the research, each with a separate dataset comprising pictures of spinach leaves with various degrees of disease severity. The federated learning system made developing precise global models easier, protecting data privacy. The findings showed that the federated learning technique uniformly enhanced clients’ performance, with Client 4 attaining the most critical performance metrics: 96.19% Precision, 95.97% Recall, 96.07% F1-Score, and 0.98 Accuracy. Our federated learning models outperformed existing distributed learning techniques and centralised learning approaches when we compared their performances. Using various averaging strategies, including macro, weighted, and micro averages, it was repeatedly shown how well the federated learning methodology classified spinach leaf illnesses. The effectiveness of this strategy implies that by allowing early diagnosis and intervention for several diseases in spinach leaves, it has the potential to increase crop output and provide food security.

Topics & Concepts