Revolutionizing Cassava Leaf Disease Detection with Federated Learning CNN and Multi-Level Severity Assessment

Abstract

Cassava is an indispensable commodity for millions of people around the globe, serving as a significant source of calories and nutrients. However, cassava leaf diseases pose a significant risk to the yield and productivity of the crop. In this study, To proposed a privacy-protecting federated learning CNN model for classifying cassava leaf disease severity levels. The model was trained on 9,675 images representing four severity levels: normal, mild, moderate, and severe. Using federated learning, the model enabled multiple farms or organizations to train the model collaboratively using their respective datasets without sharing raw images, thereby addressing data privacy concerns and enhancing the model's generalizability. The federated learning CNN model obtained an overall accuracy of 0.95 and an F1-score of 0.91875, indicating a lively performance in classifying cassava leaf diseases with varying degrees of severity. The model demonstrated exceptionally high precision and recall rates for the healthy and severe classes while revealing room for development in differentiating between the mild and moderate severity levels. Compared to traditional CNN models and other machine learning methods, the federated learning CNN model demonstrated competitive performance, addressing data privacy concerns and benefiting from diverse data sources. These results indicate that the privacy-protecting federated learning CNN model has the potential to be an effective solution for cassava leaf disease classification in real-world agricultural scenarios, ultimately leading to enhanced disease management and crop yields.

Topics & Concepts