Smart Farming: Integrated Federated Learning CNNs for Brinjal Leaf Disease Detection

This study looks into how Federated Learning (FL) and Convolutional Neural Networks (CNN) can be used in new ways to find and classify five types of Brinjal leaf diseases. The project includes cooperation with five distinct customers; each offering localized datasets covering unique symptoms of these illnesses. The way to did it combined the strength of convolutional neural networks (CNNs) for processing images with the privacy-protecting, distributed nature of FL to create a strong model that can identify many different signs of illness. The study centred on the implementation of a cutting-edge approach called Federated Averaging, which unifies the learning process across other regional datasets into a single, more generalizable model. Macro, micro, and weighted averages were used to construct the performance measures, each providing a different perspective on the model's efficacy. Analysis of the results showed wide discrepancies amongst other customers, which is to be expected given the variety of the data sets used. The overall average, weighted average, and micro average for client io_1 were 87.68%, 87.54%, and 87.54%, respectively. This showed that it could detect things in a wide range of situations. With a macro average, weighted average, and micro average of 91.16%, respectively, client io_2 did quite well in illness categorization. Client io_3 performed quite well, with a 97.20% macro average, a 97.54% weighted average, and a 97.54% micro average. Meanwhile, Client io_4 demonstrated a weighted average, micro average, and macro average of 88.22%, 88.27%, and 88.30%, respectively. However, client io_5 showed comparatively worse performance, with averages of 81.51% on the macro scale, 80.74% on the weighted scale, and 80.68% on the micro-scale.