Next-Generation Wheat Disease Monitoring: Leveraging Federated Convolutional Neural Networks for Severity Estimation

Abstract

In this study, we investigate using federated learning for the CNN model-based prediction of wheat disease severity levels. We employed safe aggregation approaches to training the CNN model on dispersed data while maintaining the privacy and security of the data. The dataset consisted of 8643 photos of wheat plants with 10 severity levels of illness. We assessed the efficacy of the federated learning CNN model using a variety of assessment measures, such as accuracy, precision, recall, F1 score, and AUC ROC, using a validation set. With an accuracy of 0.92, a precision of 0.87, a recall of 0.92, an F1 score of 0.89, and an AUC ROC of 0.95, to findings demonstrated that the federated learning CNN model performed well across all assessment criteria. The effectiveness of the federated learning CNN model and a centralized CNN model trained on the same dataset were also compared. To findings demonstrated that the federated learning CNN model could perform as well as the centralized CNN model with the best hyperparameters, with an accuracy difference of less than 0.05. To study indicates the promise of federated learning as a reliable method for estimating the severity of wheat disease using a CNN model. The ability to keep data on the device lowers the danger of data breaches and ensures user privacy. This is possible using local model training on each participating node and safe aggregation methods. To findings further highlight the significance of data distribution and thorough hyperparameter tweaking for optimum model performance. To results could stimulate more studies in this field and aid in creating federated learning strategies for machine learning across several domains.