Drone-Based Remote Sensing for Crop Health Assessment

Abstract

Unmanned aerial vehicles equipped with multispectral and thermal imaging sensors represent a transformative platform
for high-resolution crop health monitoring. This study evaluates a DJI Matrice 300 RTK drone with Micasense
RedEdge-MX multispectral and FLIR Vue Pro R thermal sensors for early detection of four crop stress conditions:
nitrogen deficiency, fungal leaf disease, water stress, and pest infestation across wheat, maize, and potato at six
experimental sites in Estonia and Switzerland over two growing seasons (2024-2025). A total of 1,847 georeferenced plot
observations were collected at eight phenological stages at 3-5 cm ground sampling distance. Random forest and CNN
classifiers were trained on computed vegetation indices (NDVI, NDRE, GNDVI, CWSI) and raw spectral data. CNN
achieved the highest overall F1-score of 0.912 for the four-class stress detection task, outperforming RF (F1=0.874).
Early detection of fungal disease was achieved 7-10 days before visible symptom appearance with 88.4% sensitivity.
Water stress detection using canopy temperature deviation achieved 91.2% accuracy. Monitoring cost of EUR 13.1 per
hectare per flight confirms operational viability for farm-scale precision agriculture deployment.
Keywords: ; ; ; NDVI; NDRE; CNN classification; Water