Unravelling Peach Leaf Disease Severity: A Federated Learning CNN Perspective

The identification and severity evaluation of peach leaf diseases using Convolutional Neural Networks (CNN) inside a federated learning (FL) framework is presented in detail in this study. Precision, recall, F1-score, and accuracy were used to measure the model's performance over various severity levels. Concerning four customers, the local data analysis showed good precision (ranging from 93.53% to 96.80%), recall (ranging from 93.59% to 97.12%), F1-score (ranging from 94.60% to 96.10%), and accuracy (ranging from 0.97 to 0.99). This outcome supports the model's reliability in categorizing various illness severity levels. The federated learning-based global analysis further supported the CNN model's effectiveness. Between the four customers, the precision values varied from 95.08% to 96.02%, recall from 94.87% to 95.85%, Fl-scores from 94.97% to 95.93%, and accuracy between 0.97 and 0.98. The efficacy of federated learning in managing and learning from decentralized data while protecting the privacy and achieving excellent performance is validated by these consistent outcomes across all criteria. The combined results highlight the model's consistent performance across all classes and are shown as macro, weighted, and micro averages. Precision levels varied between 95.94% and 96.01%, recall values between 95.08% and 95.10%, and Fl-scores between 94.96% and 94.97%, while accuracy values stayed within the range of 95.10% to 95.11% across all averages. The study's findings support the promise of federated learning with CNNs in precision agriculture, notably in identifying and evaluating peach leaf disease severity. It offers a practical, effective, and privacy-preserving method for managing decentralized data while retaining a high level of model performance. Future studies might investigate how this model can be used for different types of plants and illnesses, which would enhance the area of precision agriculture.