Transforming Crop Diagnosis: The Role of Federated Learning CNN in Cauliflower Leaf Disease Detection

This research study explored a novel way of diagnosing and categorizing cauliflower leaf diseases by integrating federated learning (FL) with convolutional neural networks. Five populations of clients (px_1 to px_5) are used to evaluate five genera of cauliflower leaf diseases involving locally distributed data. The approach that we adopted encompassed the utilization of federated averaging methods for the integration of local learnings from each client's dataset with a universal global model; further, once this was achieved, successful training of CNN models on every individual client's dataset. Thus, we used this approach to guarantee that the model could generalise better and become more robust through the efficient transfer of findings from separate datasets into a larger context. This research result can be described numerically as staggering in effectiveness. The model is compelling enough to learn from different data sources, as consistent enhancements in customer performance indicators attest. The model's performance has consistently been evaluated based on the macro averages, ranging from 91.83% in px_1 to 97.42% obtained using px_5. The weighted averages regarding class imbalance were similar to the macro averages, ranging between 91.88% in px_1 and 97.41% in px_5. These numbers demonstrate that the model worked well, regardless of how familiar a particular illness class is in the dataset. For instance, micro averages reflecting overall accuracy improved from 91.87% observed for px_1 to per cent marks ranging between and exceeding at roughly equal increments until reaching an outstanding result of % noted with px5—mirrored trends witnessed among macro-averaging.