Brinjal Leaf Pathology: An Improved Disease Classification in a Federated Learning-CNN Approach

A study about brinjal leaf disease diagnosis using federated learning using CNN is presented here. Based on the data from four clients, we can identify four levels of severity for brinjal leaf disease. As a result of the use of federated learning, it is possible to train a CNN model collaboratively while maintaining data privacy and locality, which are essential when dealing with agricultural data. A federated average algorithm was used in our approach to combine local updates of models into a global model. Using this method, the model gains experience from different data sources, thus becoming more generalized and reliable. As evidenced by enhanced model performance among clients, this approach produces tangible outcomes. There were several significant results from the analysis. The macro average of disease diagnosis accuracy through clients is 97.55% in NA_6 and 89.71% in NA_1. As a result of accounting for disease severity classes, the weighted average (Weighted_average) also displays a rising trend. Also, the Micro Average (Micro_ave) shows this trend, running from 89.55% (na_1) to 97.55% (na_6), which represents a general classification success rate. Federation learning can manage brinjal disease severity strata's intricacies and disparities successfully. For a reliable classification of medical conditions, the model must accommodate various inputs, especially among later clients. The study improves agricultural technology by addressing field infections with a practical, scalable, and privacy-preserving method. This study suggests that federated learning and CNN can be used to diagnose crop diseases, revolutionizing the practice.