Detection of maize leaf disease using machine learning

Abstract

This research investigates the development and optimization of Convolutional Neural Network (CNN) models to accurately detect maize leaf diseases. Utilizing a comprehensive dataset sourced from Kaggle, My target common maize diseases such as Maize Rust, Leaf Blight, and Gray Leaf Spot. The study aims to enhance agricultural productivity through early and precise disease identification. The dataset includes meticulously annotated images of maize leaves captured in both field and controlled environments. Various CNN architectures, including VGG-16, ResNet-50, EfficientNet, and MobileNet, are employed for image classification. These models are trained using a supervised learning approach, with key evaluation metrics such as accuracy, precision, recall, and F1-score. Our experimental results reveal that EfficientNet and ResNet-50 demonstrate superior

performance, achieving higher accuracy, precision, recall, and F1-score compared to VGG- 16 and MobileNet. EfficientNet, in particular, achieved the highest accuracy, making it the

most effective model for this application. These findings are supported by detailed statistical analyses, including confusion matrices and precision-recall curves. The implementation of this research has significant implications for society and the environment. By improving the accuracy of maize leaf disease detection, we can enhance crop management practices, reduce pesticide usage, and increase overall crop yields. This has the potential to bolster food security and provide economic benefits to farmers and the agricultural industry. The study also addresses the ethical aspects of deploying AI in agriculture, emphasizing the need for transparency, fairness, and accountability in AI systems. A sustainability plan is proposed to ensure the long-term viability of the developed models, including continuous dataset updates, model retraining, and integration with IoT technologies for real-time monitoring.