Optimizing Fenugreek Leaf Disease Detection: An Exploration of Federated Learning with CNN

Abstract

Fenugreek leaf diseases severely hamper agriculture production; thus, effective detection and classification techniques are required. This study describes a novel method for identifying six kinds of fenugreek leaf diseases using federated learning and convolutional neural networks (CNNs) across several client datasets and bypassing the need for raw data exchange to retain the Federated Learning principle of privacy preservation while ensuring accurate illness categorization. Precision, recall, F1-score, and accuracy were the four main metrics to assess the model's robustness. The model consistently performed well in individual client evaluations (part IV), with F1-scores for all classes reaching 88% in Client A, 92% in Client B, 93% in Client C, and 95% in Client D. On the globally aggregated dataset (part V), the model continued to perform well, with overall F1-scores of 91.78% for Client A, 95.67% for Client B, 93.52% for Client C, and 94.93% for Client D. Three averaging techniques—macro average, weighted average, and micro average—were used for a deeper analysis in Section VI. These averaging methods demonstrated consistently good performance across all customers, with micro and weighted averages falling between 91.98% and 95.71% and macro averages ranging from 91.80% to 95.67%. These findings highlight the model's strong performance against class imbalances across several illness classes. This study highlights the potential of CNNs and federated learning for identifying and categorising agricultural diseases. The method provides a scalable, effective, and privacy-preserving alternative with a lot of promise for increasing agricultural production, particularly for the cultivation of fenugreek.

Topics & Concepts