UAV-based multispectral imaging and machine learning for early detection of nitrogen deficiency in maize

Abstract:

Nitrogen (N) deficiency is among the most prevalent nutritional constraints limiting maize productivity across sub-Saharan Africa, yet its early and site-specific detection remains technically challenging using conventional ground-based methods. This study investigated the efficacy of unmanned aerial vehicle (UAV)-based multispectral imaging combined with machine learning algorithms for the early detection and spatial mapping of N deficiency in maize (Zea mays L.) at three critical growth stages. A field experiment was established in Oyo State, Nigeria, with five nitrogen treatment levels (N₀: 0, N₁: 30, N₂: 60, N₃: 90, N₄: 120 kg N ha⁻¹) in a randomized complete block design with three replications. A DJI Phantom 4 Multispectral UAV was deployed at V4, V8, and VT/R1 growth stages (60 m altitude, 80% image overlap) to acquire five-band imagery (blue, green, red, red-edge, near-infrared), from which eight spectral vegetation indices were derived. Three machine learning classifiers Random Forest (RF), Support Vector Machine (SVM), and Convolutional Neural Network (CNN) were trained and validated against destructive leaf N analyses as ground truth. The CNN model achieved the highest overall accuracy (91.2%) and F1-score (0.89), followed by RF (87.3%) and SVM (84.6%). The red-edge normalized difference vegetation index (NDRE) exhibited the strongest correlation with leaf N content across all growth stages (r = 0.93 at VT/R1). Treatment N₂ (60 kg N ha⁻¹) was classified as the critical threshold below which spectral-reflectance-based N deficiency signatures became detectable at the V4 stage with 88.4% accuracy. These findings demonstrate the practical potential of UAV multispectral imaging coupled with deep learning for scalable, early-stage precision nitrogen management in smallholder maize production systems across West Africa.