Adopting Federated Learning and CNN for Advanced Plant Pathology: A Case of Red Globe Grape Leaf Diseases Dissecting Severity

Abstract

This article examines a novel method for categorizing Red Globe grape leaf illnesses, supported by convolutional neural networks (CNNs) and federated learning. The goal is to aid farmers in adopting focused, efficient treatments by facilitating a more thorough and accurate diagnosis of various disease severity levels. The research uses a federated learning approach among five clients and contains five discrete severity levels of Red Globe grape leaf diseases. Because it trains a global model based on locally calculated updates and keeps data on the client device, the federated learning paradigm used here is unique in how it operates and handles data privacy issues. CNN is a powerful deep-learning method especially effective when examining visual information. The study findings' accuracy, precision, and recall qualities are encouraging for all customers. The customers' scores for local data precision, recall, F1-score, and accuracy ranged from 85.60 to 96.62%, 88.82 to 96.54%, 87.50 to 96.58%, and 0.96 to 0.99, respectively. When the model's predictions were further evaluated using global client values, accuracy scores ranged from 90.71 to 95.21%, demonstrating high predictability. A comparison study of averages was conducted to demonstrate the model's effectiveness among customers. The weighted averages ranged from 91.47 to 95.33%, while the macro-averages ranged from 90.90 to 95.25 %. The micro-averages were between 91.45 and 95.32%, consistently demonstrating the model's resilience at all severity levels. With this method, farmers, especially in India, are given a powerful tool for the early identification and treatment of grape leaf diseases, which increases production and encourages sustainable agriculture.