Cauliflower Leaf Disease: Unraveling Severity Levels with Federated Learning CNN

The present research deals with a new approach to precision agriculture, particularly the classification of cauliflower leaf disease using Federated Learning (FL) integrated into Convolutional Neural Networks. The study presents a new categorization scheme for leaf diseases that uses four severity levels based on the distribution of a dataset through 4 different clients. It aimed to exploit the decentralized nature of FL to improve data privacy while benefiting from powerful image classification potentials offered by CNNs. Two of the four clients in our study, fq1 to fq4, provided local data combined via Federated Averaging into a global model. The result analysis applied three key metrics: macro, micro, and weighted averages that portray the model's local and international performance. The results from every client indicated a steady increase in classification accuracy. The Macro Average at approximately even treatment of all the classes increased steadily from fq_1 (96.35%) until fq_4 (99.11%). Similarly, the Weighted Average that considers class representation began at 96.45% for f q_1 and ended at 99.10 % table for fq_4. The same pattern was obtained with the Micro Average in which contributions from all classes are summed up to have this average increased markedly, that is, 96.44% for fq_1 and sharply rises to a total of 99.10%% it reached its peak at time f: When computing based on percentages correctly assigned, the assumption made right before Class These averages show the overall balanced performance of the model for different severity classes regardless whether they appear more often in a dataset or not.