Bean Leaf Disease Diagnosis in the Age of Federated Learning and CNN: A Severity Analysis Approach
Shiva Mehta, Vinay Kukreja, Satvik Vats
Abstract
This study describes a unique federated learning method for classifying bean leaf disorders and determining the severity of each condition using convolutional neural networks (CNNs). The suggested methodology categorizes four distinct bean leaf diseases and rate their severity on a scale of 1 to 100% by using four separate levels. The federated learning framework allows the efficient utilization of various and dispersed agricultural data by enabling different clients to participate in model training while protecting data privacy. The federated learning technique consistently offers strong classification performance across all clients, as shown in the findings. The customers' total accuracy scores fall between 0.97 and 0.98, and their precision, recall, and F1-score scores also show excellent performance. Additionally, the aggregated model that has been internationally assessed performs consistently across a range of average metrics, such as macro, weighted, and micro average values. These findings highlight how well the federated learning framework preserves classification performance across dispersed datasets. A potential option for precision agricultural applications, the suggested federated learning strategy employing CNNs for bean leaf disease classification and severity assessment addresses data privacy issues and the need for efficient disease identification and control.