Empowering Precision Agriculture: Detecting Apple Leaf Diseases and Severity Levels with Federated Learning CNN

Abstract

The production of apples contributes significantly to the world's food security, but it also confronts significant obstacles because of illnesses that harm apple leaves. Early diagnosis and categorization are essential to effectively managing and controlling these diseases and supporting sustainable apple farming. Convolutional neural networks (CNNs) have powerful image classification capabilities. This research paper introduces a novel method to classify apple leaf diseases into four severity levels using federated learning and CNNs, aiming to harness the privacy-preserving nature and decentralized training benefits of federated learning. To replicate a federated learning environment, we created a large dataset of 8,973 labelled apple leaf photos representing a range of disease categories and severity levels. We then disseminated the data among four clients. Our decentralized training process allowed for the efficient use of various datasets and data privacy protection for our federated learning CNN model. The trial results show the efficacy of our suggested strategy, with F1 scores ranging from 93.26% to 95.94% and accuracy values for all customers between 96% and 98%. These performance measures show that the model can adequately manage data privacy and availability issues while effectively classifying apple leaf diseases. The suggested federated learning CNN model adds to the knowledge on categorizing plant diseases and provides insightful information for the next precision agriculture research. Our research highlights the potential of this approach for various applications in the agricultural domain, paving the way for more effective and privacy-preserving solutions in precision agriculture by demonstrating the viability and efficacy of using federated learning and CNNs for apple leaf disease classification.