Revolutionizing Maize Disease Management with Federated Learning CNNs: A Decentralized and Privacy-Sensitive Approach

Abstract

For sustainable maize production, reliable and prompt disease identification is crucial. Maize diseases represent a danger to world food security. In this work, utilizing decentralized and privacy-sensitive data, to constructed and evaluated a Federated Learning CNN model to identify and diagnose maize illnesses. The 9878 maize photos in to dataset, divided into training, validation, and test sets, depict a variety of diseases and conditions. To prepare the dataset for training the model, to did data pre-processing, such as picture scaling, normalization, augmentation, and label encoding. The Federated Learning CNN model outperformed conventional CNNs and other frequently used machine learning algorithms in agriculture, achieving an overall accuracy of 89.4% on the test set. The Federated Learning CNN model can precisely identify and categorize maize illnesses, as shown by the model's superior precision, recall, and F1-score values compared to other algorithms for each disease class. The CNN model from Federated Learning also showed resilience and consistency across the various disease classes, demonstrating the model's ability to adjust to local circumstances and variances in maize illnesses across multiple locations and farms. Additionally, to visualized the feature maps and activation patterns to understand the model, revealing how it generates predictions and which aspects of the maize photos are most crucial for disease diagnosis. To work underscores the significance of creating decentralized, privacy-preserving machine learning models in agriculture and shows the promise of federated learning CNNs for crop disease diagnosis and management.