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

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Maize (Zea Mays) is a vital global staple crop facing significant yield reductions due to diseases, especially in smallholder systems.
• Traditional disease detection methods are labor-intensive, subjective, and prone to errors, causing delays and substantial crop losses.

Purpose of the Study:

• To investigate the feasibility of using unmanned aerial vehicle (UAV) remote sensing and machine learning (ML) for detecting maize leaf diseases in South African smallholder farms.
• To differentiate between healthy and diseased maize plants and classify specific diseases using UAV-derived data and ML algorithms.

Main Methods:

• UAV multispectral imaging captured data, including vegetation indices (NDVI, GNDVI, NDRE) and spectral bands.
• Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) algorithms were employed for classification.
• The study focused on a smallholder farm in the Mopani District, Limpopo Province, South Africa.

Main Results:

• The Support Vector Machine (SVM) algorithm demonstrated the highest accuracy, achieving 91.73% for distinguishing healthy from diseased crops and 89.41% for classifying specific diseases.
• Southern Corn Leaf Blight was identified with the highest user's accuracy among the classified diseases.
• Phosphorus deficiency exhibited the lowest user's classification accuracy.

Conclusions:

• Integrating UAV-based multispectral imaging and ML offers a promising approach for precision agriculture in maize production.
• This technology provides timely, spatially resolved disease information, enabling targeted management strategies.
• The findings support enhanced crop loss reduction and improved food security for smallholder farmers.