Transforming Agro-Diagnostics: Banana Leaf Diseases Through Federated Learning CNNs

This paper discusses an extensive study on checking disease in banana leaves. It uses Federated Learning and Convolutional Neural Networks to find out what's wrong with them. The study uses data from five clients to examine five types of banana leaf diseases. This includes a wide variety of environmental and locational differences. The FL method, based on a federated averaging approach, helps people learn with others privately. It turns local information into a vital world model quickly and better.The central part of the study is a detailed look at how well the model works for five clients. The measures used to evaluate this include accuracy, support, and F1-score, along with precision and recall. The clients (hy_1 to hy_5) did very well. Their accuracy was always high, between 96% and 97%. The close study showed the large average was 90.41% to 93.60%, and the small average was 90.37% to 93.60%. Also, it indicated reliability in classifying diseases by giving a high level of detail with different data sources and suggesting models were very accurate for illness categorization across multiple datasets. A big part of the study involved using federated averaging to combine local data. This was important in making the worldwide model more accurate and valuable for everyone. The big, small, and weighted averages were essential to know how well the model worked. It looked at each class as they came from different places in a way that also thought about all of what we had there together. This method made sure we understood how well the model worked. It showed it could change and stay strong in different farming situations.