High resolution aerial imagery of barley over a growing season
Data files
May 19, 2023 version files 12.51 GB
Abstract
This dataset consists of unprocessed images and orthomosaic imagery of a barley field in Bozeman, Montana, collected throughout the growing season from emergence to maturity. The orthomosaics were used to develop an open-source workflow for extracting quantitative values from individual plots for downstream analysis of plant traits. This field exemplifies a challenge for plot extraction, as plots were planted with no border rows or alleys.
Methods
UAV Imagery Collection:
Data was collected using a Mavic 2 Pro drone with the integrated Hasselblad L1D-20C RGB camera at an altitude of 90 feet (27.4 m). Flights were conducted over a barley field located west of Bozeman Montana (45.676415, -111.149092). DJI GS Pro software was used on an iPad mini to create an automated flight path for imagery capture. Images were collected while hovering to minimize blurring and captured with 70% overlap along the flight path and 70% overlap between flight passes. Weather permitting, flights were timed as close to 10:00 am or 2:00 pm as possible.
Date | Number of Images | Time of Flight | Notes |
June 16 | 37 | 10.38 | |
June 21 | 49 | 11:27 | Increased number of passes for better stitching of edge plots. |
June 24 | 49 | 10:45 | |
July 01 | 49 | 10:16 | |
July 12 | 59 | 09:51 |
One-the-fly flight plan due to hardware issues. |
July 15 | 49 | 9:09 | |
July 19 | 48 | 11:20 | |
July 25 | 49 | 14:04 | |
July 27 | 49 | 14:07 | |
August 5 | 48 | 10:20 | |
August 8 | 48 | 14:44 |
OpenDroneMap was used to stitch images together and create an orthomosaic of each flight. Parameters were default except for the following arguments:
min-num-features: 4000, max-concurrency: 6, skip-3dmodel: TRUE, fast-orthophoto: TRUE, crop: 0, texturing-outlier-removal-type: gauss_damping, orthophoto-resolution: 0.125, orthophoto-compression: NONE
The minimum number of features defines the number of tie points needed to stitch each pair of images. ‘min-num-features’ was lowered from the default 8000 to 4000 to ease processing time and memory load. ‘max-concurrency’ allocates CPU cores to the stitching project. ‘skip-3Dmodel’ and ‘fast-orthophoto’ keep the stitching procedure from creating undesired files like a 3D model and digital elevation model (DEM). ‘crop’ and ‘orthophoto-compression’ maintain the imagery quality, so nothing was cropped or down sampled. ‘texturing-outlier-removal’ defines how moving objects are processed and the option ‘gauss-damping’ was chosen because it is a less aggressive approach that prioritizes images that do not include the moving object. In this image set, there were no moving objects. ‘orthophoto- resolution’ defines the final resolution of the image. A value of 0.125 was selected for this dataset as a conservative estimate of the true resolution collected by the sensor.
Field Operations:
The field was planted on April 26th, 2022, with spring barley from the S2MET population. Aggregated by Neyhart et. al. 2019, the S2MET barley population provides a representation of high-performance barley from around the United States, selected to be grown across many environments to study genotype-by-environment interactions. Lines were planted in an augmented block design including 12 blocks and four control varieties planted across all blocks. These control varieties were selected as common high-performing barley lines in the Montana region: Odyssey, Lavina, Merit 57, and Hockett. All other lines were planted once. Planting was conducted with a 6-row planter, planting two 3-row plots simultaneously in a North-South orientation. In total, 23-24 plots were planted per block, for a total of 282 plots. After emergence alleys were cut East-West to distinguish plots more easily.
Data Processing:
This dataset was used to develop an analysis workflow using QGIS and R. After stitching, imagery was loaded into QGIS. First, each image was georeferenced to the flight on June 16th using the 6 ground control points laid out over the extent of the field. Further, each band was calibrated relative to the June 16th flight image using the reflectance calibration pad (Micasense, panel serial number RP02-1622081-SC).
Once georeferenced and calibrated, plants were extracted from each image using the excess greenness index threshold (2 * Green) – Red – Blue). Next, plots were defined through a user-defined line grid overlay that was then translated into a polygon shapefile. This overlay was used to extract digital number statistics in each band, for every plot, on each flight date.
References:
- Neyhart, J.L., Sweeney, D., Sorrells, M., Kapp, C., Kephart, K.D., Sherman, J., Stockinger, E.J., Fisk, S., Hayes, P., Daba, S., Mohammadi, M., Hughes, N., Lukens, L., Barrios, P.G., Gutiérrez, L. and Smith, K.P. (2019), Registration of the S2MET Barley Mapping Population for Multi-Environment Genomewide Selection. Journal of Plant Registrations, 13: 270-280. https://doi.org/10.3198/jpr2018.06.0037crmp
Usage notes
Imagery can be previewed using any default Windows or MacOS program. The stitched orthomosaic TIFF files can also be previewed in any file viewer but are best viewed through a GIS program. The open-source GIS software QGIS is recommended.