Skip to content
Snippets Groups Projects
Select Git revision
  • cqc
  • main default protected
2 results
Blame Permalink
README.md 7.75 KiB

Imaging

Thermal Infrared Imaging

Thermal infrared imaging data were acquired using a CAT S60 smartphone equipped with a FLIR Lepton longwave infrared micro thermal camera module (https://www.catphones.com/ (accessed on 5 November 2019)).

  • Images had a resolution of 320 × 480 pixels.

  • Two images per plot were taken to cover the complete plot surface.

  • Images overlapped at the center of the plot, where a fastened piece of crumpled aluminum foil (40 cm2) and a leaf covered with petroleum jelly used as a dry reference could be included in both images.

Sampling time
  • Images were acquired at the visible inflorescence stage of development for all plots in both the FI and RI treatments in the afternoon of 46 and 47 days after sowing.
  • Images were acquired in the afternoons of 48, 49 and 50 days after sowing for all plots of blocks 2 and 5, 1 and 4, and 3 and 6, respectively.
  • At 47 days after sowing, all plots were also imaged in the late morning when the leaves were still wet with morning dew.

  • The derived temperature data were used as a wet reference.

  • Images were processed in R using the ‘Thermimage’ package [1].

  • Settings for air temperature and relative humidity at the time of imaging were obtained from the collected environment data.

  • Emissivity was set to 1 in the conversion of image data for measuring the mean temperature of the crumpled aluminum foil, which represented the reflected temperature. This was then applied together with an emissivity value of 0.96 to obtain the leaf temperature data.

  • An ImageJ macro was used to semi-automatically determine regions of interest (ROIs) in the images and corresponding temperature data for the piece of aluminum foil, the dry reference leaf and patches of sunlit, exposed soil. The temperature data of these ROIs were then excluded from the image to obtain the mean leaf temperature data.

  • Thermal index 1 (TI1) was calculated as follows:

TI1 = dTwet.m − dTm

The dTwet.m is the mean of the difference between the temperature of the wet leaves per plot at 47 days after sowing and the ambient air temperature at the time of imaging. The dTm is the difference between the mean leaf temperature per plot and the ambient air temperature at the time of imaging [2,3].

References:
[1] Tattersall, G.J. Thermimage: Thermal Image Analysis. Available online: https://CRAN.R-project.org/package=Thermimage
[2] Maes, W.H.; Steppe, K. Estimating evapotranspiration and drought stress with ground-based thermal remote sensing in agriculture: A review. J. Exp. Bot. 2012, 63, 4671–4712.
[3] Perich, G.; Hund, A.; Anderegg, J.; Roth, L.; Boer, M.P.; Walter, A.; Liebisch, F.; Aasen, H. Assessment of multi-image unmanned aerial vehicle based high-throughput field phenotyping of canopy temperature. Front. Plant Sci. 2020, 11, 150.

Hyperspectral Imaging

Hyperspectral image data were acquired using a Specim IQ (Specim Ltd., Oulu, Finland), a handheld push broom camera system with integrated operating system and controls [1]. The Specim IQ measures reflectance in the visible and near-infrared, i.e., from 400 to 1000 nm, with a spectral resolution (FWHM) of 7 nm, 204 spectral bands, and a spatial resolution of 512 × 512 pixels2. The camera was mounted on a tripod at a height that allowed a complete individual plot to be captured in the image.

Sampling time
  • The plots of blocks 1 and 4, and 2 and 5 were imaged at 49 and 48 DAS, respectively, between 15:00 and 16:00.
  • The plots of blocks 3 and 6 were imaged at 49 DAS between 16:00 and 17:00.
Repetitions

As blocks represented differences in the date and time of imaging, they were referred to as repetitions:

  • Blocks 1 and 4 assigned to repetition 1,
  • Blocks 2 and 5 to repetition 2, and
  • Blocks 3 and 6 to repetition 3.

  • Plots were always captured in treatment pairs (RI after FI or vice versa).

  • Each dataset contained a white reference tile, which was imaged simultaneously for data calibration. A dark reference, representing sensor noise without incoming light, was recorded automatically before each capture.

  • Upon image data acquisition, the Specim IQ integrated software allows for the selection of the white reference tile in the image based on its high reflectance values, in addition to automated calibration to obtain relative reflectance data. However, we noted that the white reference tile itself was not selected alone in some images, as other elements with high reflectance were present, such as pieces of crumpled aluminum foil (used for the measurement of stem water potential and thermal imaging). The calibration procedure was therefore redone in R after threshold-based selection of the white reference tile pixels using an ImageJ macro. All other hyperspectral data processing and analysis steps were also executed in R.

  • Plant pixels were segmented from the background using the Normalized Difference Vegetation Index (NDVI, Table S1) and a threshold level, which also excluded inflorescences and specular reflection. Shaded background and shaded plant parts with low reflectance were removed using a threshold in near-infrared (838 nm) and green (554 nm) wavelengths, respectively. Spectra were smoothed on the pixel level using the Savitzky–Golay smoothing filter [2] with a third order polynomial and a window size of 11 using the R package ‘prospectr’ [3].

  • A total of 41 published vegetation indices (VIs, Table S1) were calculated. By means of a cluster analysis, genotypes were grouped based on the similarity of VI data within the FI and RI treatment. Agglomerative hierarchical clustering was applied on the scaled VI mean observations for repetitions 1 to 3 using the ‘agnes’ function of the R package ‘cluster’ [4]. The trees were cut at five clusters. Pearson correlation coefficients were calculated to describe the linear relationships between traits measured at the visible inflorescence stage, plant morphology and performance traits measured at harvest, and VIs. In addition, differences in relative reflectance between genotypes and treatments, independent of VIs, were analyzed for a selection of wavelengths and wavelength bands. A selection was used because of the high degree of correlation or collinearity in the relative reflectance of mostly adjacent wavelengths. A Pearson correlation coefficient was calculated between the relative reflectance of all wavelengths. A threshold of 0.8 was then applied to split up the wavelength range in groups of high correlation. One wavelength was selected for further analysis per group. This yielded five wavelengths, 476 nm, 554 nm, 616 nm, 679 nm and 724 nm, in the blue, green, orange, red and red-edge regions of the spectrum, respectively. Furthermore, reflectance in the near-infrared region (NIR) was averaged and included in the selection.

References:

[1] Behmann, J.; Acebron, K.; Emin, D.; Bennertz, S.; Matsubara, S.; Thomas, S.; Bohnenkamp, D.; Kuska, M.T.; Jussila, J.; Salo, H.; et al. Specim IQ: Evaluation of a new, miniaturized handheld hyperspectral camera and its application for plant phenotyping and disease detection. Sensors 2018, 18, 441.
[2] Savitzky, A.; Golay, M.J.E. Smoothing and differentiation of data by simplified least squares procedures. Anal. Chem. 1964, 36, 1627–1639.
[3] Stevens, A.; Ramirez-Lopez, L. Miscellaneous Functions for Processing and Sample Selection of Vis-NIR Diffuse Reflectance Data. Available online: https://github.com/l-ramirez-lopez/prospectr (accessed on 30 November 2020).
[4] Maechler, M.; Rousseeuw, P.; Struyf, A.; Hubert, M.; Hornik, K.; Studer, M.; Roudier, P.; Gonzalez, J.; Kozlowski, K.; Schubert, E.; et al. “Finding Groups in Data”: Cluster Analysis Extended Rousseeuw et al. Available online: https://cran.r-project.org/web/packages/cluster/cluster.pdf (accessed on 26 February 2021).