A New Era in AgriTech: Federated Learning CNN for Jute Leaf Disease Identification

A novel method for classifying disease severity in jute leaves is developed using federated learning and a convolutional neural network (CNN). For this research, four clients with different severity levels of jute leaf disease were used. It is then possible to apply federation learning to split gathered data into local and global segments while maintaining privacy simultaneously. The federated fusion modelling method can also produce global models that are powerful. FedAvg is an algorithm that efficiently combines local updates into a worldwide model, which we use in this approach. Macro Averages, Micro Averages, and Weighted Averages are used to evaluate this approach. The result analysis demonstrates the skill of converting local data into global access to combat jute leaf disease.Using numerical evaluation criteria, four clients were classified into a wide range of diseases based on their evaluation criteria. A wide range of Macro Average values can be observed between severity classes, with 73.41 % to 90.72 % being the maximum. In the target population, classes more prevalent than others accounted for 73.12 % to 90.73 % of the weighted average, showing how well the model performed on those classes. This model is helpful in categorizing individual instances into appropriate groups based on the Micro Average, which represents the combined contribution of all classes. Additionally, this study shows that federated learning and CNNs can solve real-world agricultural technology problems. A precision agriculture application based on jute leaf disease severity can be developed based on the findings of this study. Jute is a staple crop in many areas, making the results of this study crucial for improving crop quality and yield.