Soybean iron deficiency chlorosis high-throughput phenotyping using an unmanned aircraft system

Table 1 Pipeline for image capture and analysis for iron deficiency chlorosis assessment

Category	Step	Details
UAS image collection	Set up UAS	DJI Inspire 1 with Sentera Double 4 K sensor
	Prepare flight path	AgVault mobile app or Pix4D capture app
	Fly UAS for data collection	70% image overlap, 61 m altitude
Image orthomosaic using Pix4D	Initial processing	Key points extraction and matching, camera model optimization, geolocation
	Point cloud and mesh	Point densification and 3D textured mesh creation, insert ground control points
	Digital surface model, orthomosaic, and index	Creation of digital surface model, Orthomosaic, reflectance map, and index map
Image processing	Plant and soil classification	k-means clustering and recode to plant and soil
	Green, yellow, brown pixel classification	k-means clustering on masked canopy and recode to green, yellow, brown
	Neural network/random forest with ground data	Subset into training and validation sets, ground-based data is response variable and green, yellow, brown pixel counts are used as features

The flight path is set up in Pix4D capture with 70% overlap of images. Individual photos are orthomosaiced in Pix4D and k-means clustering is used to mask the plants from the soil background. An additional classification of green, yellow, and brown pixels is performed on the plant objects. In QGIS, plots are defined, and the proportions of green, yellow, and brown pixels are extracted from each plot. Finally, predictions are made to correlate these three features with ground based visual score estimates rated on a 1–5 scoring system

ISSN: 1746-4811