Crop Disease Prediction Using Deep Learning in a Federated Learning Environment: Ensuring Data Privacy and Agricultural Sustainability

Abstract

Introduction: The traditional approach may not be suitable for crop disease prediction because there may be many privacy and security concerns. We have introduced the Federated learning approach, which does not allow the sharing of complete data across multiple devices. This research approach is the combination of federated learning (FL) with deep neural networks to increase crop disease prediction as well as data privacy. Objective: The main objective of our article is to develop a robust and privacy-preserving crop disease prediction model that is combined with federated learning and deep neural networks within a federated learning environment. This strategy improves prediction accuracy while maintaining data sovereignty by storing data locally and securely. Material/Method: Initially we deployed the IoT sensors (in different parts of the land, and then we collected the data continuously to monitor the environmental conditions. By taking the image data and the collected sensor data, the model trained and understood the factors for crop health. The image data will be preprocessed through resizing, normalization augmentation, etc. We have used deep neural network classifiers like CNN, RNN, LSTM, and GRU. The federated learning technique is implemented. A global model is created to update the aggregate of the locally trained model. Result: Unlike traditional centralized approaches, our method maintains data sovereignty by keeping data local while gaining global insights from a federated paradigm. This is especially important in places where data privacy regulations prevent data exchange. Furthermore, our approach combines IoT sensor data with deep learning frameworks to provide a more comprehensive understanding of crop health. We have estimated accuracy, precision, recall, and F1-score AUC-ROC and PR-AUC to evaluate the model’s performance on the real-time dataset. Our experimental results reveal that CNN achieved 99% training accuracy and 94% testing accuracy as compared to other deep learning techniques for crop disease prediction with a federated learning environment. A marginal difference between training and testing accuracy, i.e. 0.05%, indicates that the model is robust.

Topics & Concepts