Evolving Agritech: Implementing Federated Learning & CNN for Parsley Leaf Disease Detection

Abstract

This study investigates a unique approach that combines federated learning with a Convolutional Neural Network (CNN) to identify illnesses of the parsley leaf. It illustrates the effectiveness of using decentralised data among several clients to produce highly effective, reliable models while protecting data privacy. The empirical results show improved model performance for six illness classes (Class_A through Class_F) and four customers (Cl_a, Cl_b, Cl_c, and Cl_d), with precision, recall, F1-score, and accuracy typically above 95%. For instance, in the regional analysis, Client B attained impressive accuracy and recall of 97.39% and 97.85% for Class_F, suggesting little to no false positives or negatives for this class. In the overall study, the federated learning average results show similarly impressive results, such as Client B, which produced precision, recall, F1-score, and accuracy values of 96.45%, 96.31%, and 99%. The little differences between local and global metrics demonstrate federated learning's resilience, which successfully maintains high performance in the face of data decentralisation. Macro, micro, and weighted averages, which provide subtle hints about the model's performance across various classes and data distributions, are also explored in this study. For example, Client D showed a macro-average precision of 95.47%, a micro-average precision of 95.35%, a weighted average precision of 95.41%, and an accuracy of 99%, demonstrating the model's consistency in effectiveness under various circumstances. The findings show the value of federated learning as a decentralised, privacy-preserving, and high-performing disease detection method in parsley crops. Thus, this study contributes substantially to creating tech-driven, sustainable solutions for agriculture, potentially benefiting farmers worldwide, especially those in India.

Topics & Concepts