Advancing Agricultural Practices: Federated Learning-based CNN for Mango Leaf Disease Detection

Abstract

This study offers a new method for identifying and categorizing mango leaf illnesses using a Convolutional Neural Network (CNN) model based on federated learning. Diseases that affect mango leaves significantly reduce crop output and quality and threaten farmers' ability to make a living. Early and precise disease identification is essential for successful care and preventing these illnesses' spread. We examine the efficacy of our suggested model on four distinct customers while concentrating on the five disease classifications Healthy, Anthracnose, Powdery Mildew, Leaf Spot, and Leaf Curl. With precision values ranging from 93.33% to 96.01%, recall values ranging from 90.59% to 97.45%, F1-scores ranging from 92.64% to 96.10%, and accuracy values between 97% and 98%, the model exhibits solid performance across all clients and illness classes. The macro, weighted, and micro averages, with macro averages ranging from 93.18% to 94.97%, weighted averages ranging from 93.26% to 95.08%, and micro averages ranging from 93.26% to 95.08%, further highlight the model's consistent performance across various clients and disease classes. The federated learning-based CNN model successfully addresses the difficulties farmers encounter in identifying and controlling mango leaf diseases, resulting in more effective and long-lasting agricultural practices. The methodology protects data privacy using federated learning, allowing clients to cooperate and gain from shared learning without jeopardizing their data. Our research helps the agriculture industry create more sophisticated and precise disease detection techniques, fostering better crop management and increased food security.

Topics & Concepts