Beyond Orthomosaics: A comparison of orthomosaic versus original aerial images for high-throughput phenotyping
Date
2023-06-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
Type
Thesis
Degree Level
Masters
Abstract
Imagery from Unoccupied Aerial Vehicles (UAVs) is frequently used for remote crop assessment and phenotyping in agricultural research. In particular, spectral indices such as the Normalized Difference Vegetative Index (NDVI), are commonly employed to estimate traits such as the health, growth stage, and biomass in crops. In virtually all such studies, many overlapping images from a UAV flight are first stitched into a single orthomosaic image, and then spectral indices are derived from the orthomosaic. This method necessarily discards (or aggregates) much of the original pixel information. For example, a single plot may appear in 10 or more individual images from the UAV, but appear only once in the final orthomosaic. In this research, we show that an NDVI value extracted from individual calibrated images of the same plot can deviate by up to 0.05 from the NDVI value of the corresponding orthomosaic segment. This could have important consequences for fine-grained comparison of indices. To address this problem, we propose alternative approaches to estimating indices, each of which more directly incorporates all of the individual UAV images. We evaluate these approaches, and compare them with the orthomosaic approach, by analyzing weekly index values from plant breeding experiments for lentil, wheat, and canola crops in comparison to relevant manually-measured phenotypes.
Description
Keywords
orthomosaic, UAV, agriculture, imagery, NDVI, unstitching, plot segmentation
Citation
Degree
Master of Science (M.Sc.)
Department
Computer Science
Program
Computer Science