Apple Leaf Disease Recognition: A Robust Federated Learning CNN Methodology

Abstract

Apple leaf diseases threaten apple orchard sustain ability and production worldwide. Accurate and early identification is essential for the successful care and control of many disorders. This article suggests a convolutional neural network (CNN) model for classifying apple leaf diseases based on federated learning. Using a dataset of 7,832 photos with a resolution of 224×224 pixels and five distinct classes of apple leaf disease, the model is intended to identify these diseases. Assuring data confidentiality and privacy, the federated learning architecture enables decentralized model training across several clients. The suggested model was tested on four customers, and the results showed that the model's precision, recall, and accuracy ranged from 93.51% to 94.92%, 93.22% to 94.72%, 93.33% to 94.81%, and 0.97 to 0.98, respectively. These findings show how the federated learning CNN model outperforms alternative deep learning architectures and conventional machine learning techniques in identifying apple leaf diseases. The model generalizes well to different data distributions based on consistent client performance, which qualifies it for use in practical applications. Future research should address federated learning's drawbacks and restrictions, such as heterogeneous data distribution, model convergence, and resource limits. Additionally, investigating future uses in agriculture and plant disease control may result in a better judgment and more environmentally friendly agricultural methods.