A Federated Learning CNN Approach for Tomato Leaf Disease with Severity Analysis

Abstract

This research includes four disease levels and one healthy level in a federated learning Convolutional Neural Network (CNN) model for detecting and categorizing tomato leaf illnesses across five severity levels. Data from four customers were used to analyze the model, with each client’s performance measures for precision, recall, F1-score, and accuracy being reviewed. The obtained results show that the model consistently generated results of good quality, with an accuracy range from 96% to 98% across all severity levels. Client 4 had the greatest performance in Class 1 (healthy), but Clients 2 and 3 achieved the same results in every class, suggesting that the datasets or data distribution were comparable. Class 5 (disease 4) had the highest recall for all clients, demonstrating the model’s skill in identifying this illness. An accuracy of 97% was consistently shown when the model was assessed using locally averaged global values of parameters across four clients. Additionally, this study has contrasted the model’s performance across four clients using the macro average, weighted average, and micro average averaging techniques. These results highlighted the model’s potential usefulness in agricultural settings by demonstrating consistent performance utilizing averaging techniques across all customers. While protecting data privacy, the federated learning CNN model accurately identifies tomato leaf diseases at various severity levels, allowing for targeted agricultural practices and interventions to reduce yield loss and improve fruit quality.