UAV-based multispectral imaging and machine learning for detecting and mapping maize leaf diseases in smallholder farms.

Abstract

Maize (Zea Mays) is one of the world's most important staple crops, providing food for humans and feed for livestock. However, its production is threatened by a range of stresses, including crop diseases, which significantly reduce yields, particularly in smallholder farming systems. Traditional disease detection methods, such as visual inspection, are often labour-intensive, subjective, and prone to error, leading to delayed interventions and widespread crop losses. This study uses unmanned aerial vehicle (UAV) remote sensing and machine learning (ML) to investigate the feasibility of detecting maize leaf diseases in a smallholder farm located in the Mopani District of Limpopo Province, South Africa. UAV-derived vegetation indices including NDVI, GNDVI, and NDRE were combined with UAV multispectral bands and the three ML algorithms, namely - support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost), to first distinguish healthy from diseased plants and then to classify specific maize diseases. The SVM algorithm achieved the highest accuracy in both, distinguishing healthy and diseased crops from other land cover classes (91.73%) and in distinguishing specific diseases (89.41%). Among the diseases identified, Southern Corn Leaf Blight was classified with the highest user's accuracy, while phosphorus deficiency had the lowest user's classification accuracy. The results demonstrate the potential of integrating UAV-based multispectral imaging and ML for precision agriculture by providing timely, spatially detailed disease information that enables targeted management practices, reducing crop losses and enhancing food security for smallholder farmers.

Plant phenotyping relevance

UAVマルチスペクトル画像と機械学習を用いて、トウモロコシの健全・罹病状態および具体的な葉病害を推定する手法が研究の中心であり、植物病害状態のフェノタイピングに該当する。

Code and data availability

The article describes UAV multispectral imagery, field-collected ground truth data, and RF/SVM/XGBoost classification in R, but contains no data availability statement, no public repository deposit, and no author code URL. All repository-like URLs in the text are citations to prior work or generic references (e.g., CRR

No evidence-backed public reproduction asset is currently recorded.