AI in Agriculture: Federated Learning CNNs for Sunflower Leaf Disease Diagnosis

The research employs the combination of CNNs and FL architecture to offer a new approach to identifying and classifying sunflower leaf diseases. Four clients are used in the studies, and they present four levels of severity for diseases that impact sunflower leaves. Each client's data is unique for constructing a valid model that accounts for the wide range of environmental factors and illness presentations. The core component of our methodology is federated learning, which ensures data privacy while reducing the need for centralized storage and allowing a CNN model to be trained cooperatively by multiple clients. The model's performance is measured using accuracy, precision-recall, F1-score, and support metrics. The results analysis reveals that the approach works for all four customers. However, the macroaverage scores consistently demonstrated a capacity to diagnose various levels of severity between 90.35% and 87% for all classes. The weighted averages, varying from 90.41% to 94.85%, including the class support, are illustrated below. On the same note, microaverages that work well for class-imbalanced datasets revealed outstanding ranges between 90.41% and 94.83
^%
. These results show that the model's performance remains accurate for various class frequencies and distributions. Additionally, federated averaging was critical for transforming the local data analysis into a unified global model. This can be seen in converting local data to global data, where federated averaging enabled combining numerous learnings from different locations into a comprehensive picture of the sunflower leaf disease level worldwide. The statistical and experimental results unambiguously reveal a favourable impact of federated learning with CNNs for agricultural disease detection and classification.