Plant disease detection and classification: based on machine learning and Eig(Hess)-co-occurrence histograms of oriented gradients

Abstract

Agricultural districts provide high-quality food and contribute substantially to economic growth and population support. However, plant diseases can directly reduce food production and threaten species diversity. The use of precise, automated detection techniques for early disease identification can improve food quality and mitigate economic losses. Over the past decade, numerous methods have been proposed for plant disease classification, and in recent years the focus has shifted toward deep learning approaches because of their outstanding performance. In this study, we employ the Eig(Hess)-co-occurrence histograms of oriented gradients (CoHOG) descriptor alongside pre-trained machine-learning models to accurately identify various plant diseases. We apply principal component analysis (PCA) for dimensionality reduction, thereby enhancing computational efficiency and overall model performance. Our experiments were conducted on the popular PlantVillage database, which contains 54,305 images across 38 disease classes. We evaluate model performance using classification accuracy, sensitivity, specificity, and F1-score, and we perform a comparative analysis against state-of-the-art methods. The findings indicate that the approach we proposed achieves up to 99.83% accuracy, outperforming existing models. Additionally, we test the robustness of our method under various conditions to highlight its potential for real-world agricultural applications.