Agriculture Breakthrough: Federated ConvNets for Unprecedented Maize Disease Detection and Severity Estimation

Abstract

Maize is an important crop all over the globe, producing food, animal feed, and raw materials for various businesses. However, maize production is jeopardized by many diseases that may reduce output and decrease grain quality. Traditional disease detection and control procedures have had slight effectiveness, prompting the development of new tactics to assure long-term maize production. Machine learning and artificial intelligence (AI) developments, notably convolutional neural networks (CNNs), have shown promise in crop disease identification and diagnosis. Nonetheless, the decentralized structure of agriculture poses difficulties for traditional CNNs, which need centralized data for training. Federated learning (FL), a distributed learning paradigm, overcomes these issues by allowing model training on decentralized, private data. In this work, to explore the efficacy of an FL CNN model in recognizing and diagnosing maize illnesses using a dataset of 9,586 maize leaf pictures from diverse farms and locations with varied disease categories and severity levels. The suggested FL CNN model outperformed traditional CNNs and other machine learning approaches typically used in agriculture regarding accuracy, precision, recall, and F1-score in identifying and categorizing maize illnesses and their severity levels. The FL method can increase disease diagnosis, adjust to geographical changes, and protect data privacy, making it a viable option for real-world maize disease detection and management.