Pioneering a Beetroot Disease Diagnosis with Federated Learning and CNN
Shiva Mehta, Vinay Kukreja, Satvik Vats
Abstract
To identify six types of beetroot leaf diseases across four clients, this research article thoroughly investigates the implementation of federated learning utilizing convolutional neural networks (CNNs). This work dramatically improves the preventative measures used to control these illnesses by utilizing the potential of decentralized learning to simplify the identification and categorization of leaf diseases from photos obtained locally at diverse customers' locations. With the data staying at the local customer locations, the offered technique addresses any possible privacy issues due to its one-of-a-kind, distinctive strength. All assessment measures for our CNN model—precision, recall, F1-score, and accuracy—showed consistently strong performance. The local data from each customer showed impressive outcomes in the client-specific result analysis. For instance, in Cl_q, the model's precision, recall, F1-score, and accuracy scores reached 95.59%, 96.03%, 95.61%, and 0.98, respectively, proving the model's dependability.Furthermore, the four customers' global averages performed very well across macro, weighted, and micro averages, with accuracy in the early to mid-nineties. For example, the micro standards varied between 92.56% (Cl_x) and 94.66% (Cl_q), whereas the macro averages among customers ranged from 92.59% (Cl_x) to 94.68% (Cl_q). The research's incremental performance gain from Cl_x to Cl_q also illustrates the influence of data variety in a federated learning environment. This paper substantially contributes to the field by outlining a practical, effective, and privacy-preserving approach for identifying beetroot leaf diseases. As a result, this study provides a revolutionary method for identifying plant diseases, significantly reducing the time required for diagnosis, reducing human error, and enabling prompt action.