Decentralized Detection of Cassava Leaf Diseases: A Federated Convolutional Neural Network Solution

Abstract

Cassava is a vital sustenance crop for millions of individuals around the globe, especially in developing nations. However, cassava production is threatened by various leaf maladies that can drastically reduce productivity. Effective management and prevention of these diseases require early detection and diagnosis. This study devised a federated learning-based Convolutional Neural Network (CNN) model for the classification of five cassava leaf disease classes: Cassava Bacterial Blight (CBB), Cassava Brown Streak Disease (CBSD), Cassava Green Mottle (CGM), Cassava Mosaic Disease (CMD), and Healthy. The proposed model showed promising performance on a dataset of 9,465 images, with precision values between 0.74 and 0.85, recall values between 0.70 and 0.88, and overall accuracy of 78.6%. This study's federated learning approach offers several advantages, including data privacy, scalability, and continuous learning, making it suitable for real-world cassava leaf disease classification applications. In addition, this study demonstrates the potential for utilizing advanced machine learning techniques in the agricultural sector to improve early disease detection and support sustainable farming practices.