This codebook.txt file was generated on 2018/04/12 by Ali Moghimi ------------------- GENERAL INFORMATION ------------------- 1. Hyperspectral image dataset for salt stress phenotyping of wheat 2. Author Information Name: Ali Moghimi Institution: University of Minnesota - Twin Cities Address: Department of Bioproducts and Biosystems Engineering Email: moghi005@umn.edu Name: Ce Yang Institution: University of Minnesota - Twin Cities Address: Department of Bioproducts and Biosystems Engineering Email: ceyang@umn.edu Name: Marisa E. Miller Institution: USDA-ARS Address: Cereal Disease Laboratory Email: marisa.e.miller@ars.usda.gov Name: Shahryar Kianian Institution: USDA-ARS Address: Cereal Disease Laboratory Email: shahryar.kianian@ars.usda.gov Name: Peter Marchetto Institution: University of Minnesota - Twin Cities Address: Department of Bioproducts and Biosystems Engineering Email: pmarchet@umn.edu 3. Date of data collection: 2016/05/12 4. Geographic location of data collection: USDA Cereal Disease Lab Located in: University of Minnesota - St Paul Campus Address: 1551 Lindig St, St Paul, MN 55108 5. Information about funding sources that supported the collection of the data: United States Department of Agriculture-Agricultural Research Service the National Science Foundation (IOS 1025881 and IOS 1361554) Minnesota Agricultural Experiment Station -------------------------- SHARING/ACCESS INFORMATION -------------------------- 1. Licenses/restrictions placed on the data: CC0 1.0 Universal https://creativecommons.org/publicdomain/zero/1.0/ 2. Links to publications that cite or use the data: 3. Links to other publicly accessible locations of the data: N/A 4. Links/relationships to ancillary data sets: N/A 5. Was data derived from another source? N/A 6. Recommended citation for the data: Moghimi, Ali; Yang, Ce. (2018). Hyperspectral image dataset for salt stress phenotyping of wheat. Retrieved from the Data Repository for the University of Minnesota, http://hdl.handle.net/11299/195720. --------------------- DATA & FILE OVERVIEW --------------------- 1. File List A. Filename: All Hyperspectral images Short description: Hyperspectral images are in .bip (band interleaved by pixel) format for each wheat line as follows: a. Kharchia b. Chinese Spring (CS) c. Aegilops columnaris KU11-2 (CS) (abbreviated co(CS) hereafter) d. Ae. speltoides aucheri KU2201B (CS) (abbreviated sp(CS) hereafter) Further Notes: In all images, upper tank contains salt stress plants, and the tank at the bottom contains control plants. Per each hyperspectral image, there is a corresponding header file containing the image metadata. Open hyperspectral images: ENVI ERDAS IMAGINE Spectronon (free software available at https://downloads.resonon.com/) Programming language such as MATLAB or PYTHON B. Filename: Kharchia.mat, CS.mat, co(CS).mat, and sp(CS).mat files Short description: These .mat files contain the segmented vegetation pixels and their correponding class labels (control or salt) for all four wheat lines. These files can be imported to Matlab, Octave, and Python. a. rows: pixels (samples); columns(1:215): wavelengths (features); last column: class labels (control or salt). b. class 1 denotes the control class and class 0 denotes the salt class. C. Filename: Kharchia.txt, CS.txt, co(CS).txt, and sp(CS).txt files Short description: These .txt files (ASCII-delimited) contain the segmented vegetation pixels and their correponding class labels (control or salt) for all four wheat lines (same as .mat files). D. Filename: Kharchia.dat, CS.dat, co(CS).dat, and sp(CS).dat files Short description: These .dat files (comma-separated) contain the segmented vegetation pixels and their correponding class labels (control or salt) for all four wheat lines (same as .mat files). E. Filename: wavelength.mat, wavelength.txt, and wavelength.dat files Short description: These files are the same, containing 215 wavelengths (features) scanned by the camera. F. Filename: Kharchia.tif, CS.tif, co(CS).tif, and sp(CS).tif files Short description: These .tif files are the RGB representation of the hyperspectral images. 2. Relationship between files: .mat/.dat/.txt files contain segmentd vegetation pixels from the hyperspectral images (.bip files). 3. Additional related data collected that was not included in the current data package: N/A 4. Are there multiple versions of the dataset? No -------------------------- METHODOLOGICAL INFORMATION -------------------------- 1. Description of methods used for collection/generation of data: Sample preparation and conventional phenotyping for salt tolerance screening To develop analytical methods for analysis of hyperspectral images, four bread wheat (Triticum aestivum) lines were selected with varying levels of salt tolerance. The cultivar Kharchia was included as it is historically known to maintain a stable harvest index and yield well in high salt conditions (Munns et al., 2006; Schachtman et al., 1992), and the salt-sensitive cultivar Chinese Spring (CS) was selected as well (Zhang et al., 2016). Two additional “unknown” lines were selected for screening from a set of wheat alloplasmic lines developed in Japan (Tsunewaki, Wang, & Matsuoka, 1996; Tsunewaki, Wang, & Matsuoka, 2002). Alloplasmic lines are created by substitution backcrossing to replace the cytoplasmic genomes of one species (in this case, bread wheat) with those of another (in this case, wild wheat relatives) while maintaining the original nuclear genome background, and have shown promise for improving stress tolerance and other developmental traits (Liberatore et al., 2016). The two alloplasmic lines selected were Aegilops columnaris KU11-2 (CS) (abbreviated co(CS) hereafter) and Ae. speltoides aucheri KU2201B (CS) (abbreviated sp(CS) hereafter) with the cytoplasmic genome type preceding the nuclear genome background, which in this case is Chinese Spring (CS). Screening was performed in a hydroponic system in a Conviron growth chamber to ensure uniform conditions. Hydroponic systems are commonly used to screen plants for salt tolerance, including wheat. In all experiments, growth conditions in the Conviron were set at 22ºC during light conditions and 18ºC during the dark, 16h photoperiod, 375 μmol m-2 s-1 light intensity, and 50% relative humidity. Three hydroponic tanks were used per treatment (control treatment: 0 mM NaCl and salt treatment: 200 mM NaCl). Each hydroponic tank contained a grid of 16 Cone-tainers (Ray Leach brand) filled with perlite. Within each tank, there were four genotypes each with four individual replicates. For each treatment (salt or control), there were three replicate tanks, for a total of 48 (3 replications x 4 Cone-tainers x 4 genotypes) Cone-tainers for each treatment. The grid was placed into a tank just large enough to hold the grid, and 20L of hydroponic solution was used per tank. Genotypes were randomly assigned to positions in each cone-tainer grid using the sample and matrix functions in R (version 3.4.0). Aeration was supplied to each tank with an aquarium pump and two large airstones (at either end of the tank). Lines were transplanted into the tanks, and the lights and aeration were switched on 24 hours after transplanting. When leaf 1 emerged, ¼ strength Hoagland’s solution (PhytoTech H353) was added, and the pH was adjusted to 6.5. When leaf 2 emerged, the Hoagland’s was increased to ½ strength in all tanks and CaCl2 was added to the tanks destined for salt treatment in a 15:1 molar ratio of NaCl to CaCl2. When leaf 4 emerged, salt was added to the salt tanks gradually over 2 days to reach a final concentration of 200 mM. The water level and pH (to 6.5) were adjusted 3 times per week throughout the experiment. Two weeks after salt treatment was applied, both aerial and root biomass was harvested separately for each individual plant. Plant matter was dried at 60-65ºC for 4 days and then weighed. Dry weight data were analyzed in R (version 3.4.0) using ANOVA (car package, version 2.1-5) and linear mixed-effect modeling (nlme package, version 3.1-120). For linear mixed-effect modeling analysis, dry weight was considered as the response in the analysis, salt level and genotypes were considered as fixed effects, and tank number was considered as a random effect. Model results were identical if tank position was considered as a nested-random effect of tank number, thus the results with tank number were used as the only random effect. To compare the response of the alloplasmics to the response of the euplasmic parents when the salt level is changed, the coefficient estimates of the lme model were examined. Hyperspectral image acquisition All tanks were transferred from the Conviron to greenhouse to take hyperspectral images under natural light conditions. Each hyperspectral image contained both salt (the tank at the bottom in images) and control (the tank at the top in the images) tanks of a single wheat line. Image acquisition was done approximately 24 hours after salt application when there were no visual symptoms. To reduce the effects of sun angle and shade, images were captured close to noon (i.e., between 11:00 and 13:00 local time). A push broom (along-track scanner) hyperspectral camera (PIKA II, Resonon, Inc., Bozeman, MT 59715, USA) was used for image acquisition, which required constant movement during image capture for two-dimensional spatial information to be accurate. A glide gear slider was used to mount the camera on a horizontal bar. A Dayton DC gearmotor (model: 2L008, Dayton electric Mfg Co. Lake Forest, IL 60045, USA) was utilized to move the slider along at a set speed, with the camera oriented to face downwards. All of this was done as per Moghimi et al. (2017). The camera scanned over 240 spectral channels ranging from 400 nm to 900 nm with a spectral resolution of about 2.1 nm and captured 640 pixels in the cross-track direction (i.e., perpendicular to the direction of camera motion). The number of pixels in the along track direction was set to 2000 to assure both control and salt tanks of each line were captured in a single image. Therefore, pixel size of each hyperspectral image, also known as hyperspectral data cube, was 2000×640×240. The frame rate of the camera was adjusted based on the field of view (FOV), the distance between lens and target, and the speed of the camera motion as described by Moghimi et al. (2017). The field of view (FOV) of the camera lens was 33 degrees, and the distance between the target and lens was about one meter. The speed of the camera was set to 0.025 m/s, thus the calculated frame rate was 27 frames per second to obtain square pixels (aspect ratio of 1:1). Gain and exposure time were adjusted appropriately based on light conditions to avoid over-exposure while taking advantage of the full dynamic range (12 bits). Image preprocessing Radiometric calibration Raw images were radiometrically calibrated to account for non-uniform spatial and spectral responses of the sensor due to variability in gain and offset of each detector. Raw digital numbers (DNs) were converted to radiance (Wm^(-2) 〖sr〗^(-1) 〖nm〗^(-1)) using the radiometric calibration file provided by the camera manufacturer. Radiance was then converted to reflectance to normalize image data based on incoming solar irradiance so objects could be compared more objectively across images and across capture dates. A Spectralon panel (Labsphere, Inc., North Sutton, NH, USA) was placed in each image and was used as a reference to convert from radiance to reflectance. Spectralon reflects ~99.7% of incident light equally in all directions regardless of the illuminated light angle. Radiometric conversions were performed using Spectronon Pro software (Resonon, Inc., Bozeman, MT, USA). Noisy band removal Due to high noise, the first and last five bands were removed prior to any analysis. In addition, spectral bands from 753 to 766 nm and also from 813 to 827 nm were disregarded since they were noisy bands near the O2 (~760 nm) and H2O (820 nm) absorption regions. Following band removal, 215 of 240 bands were used for analysis. Subsequent analyses were performed using MATLAB R2017a (MathWorks, Inc., Natick, MA, USA). Vegetation mask Segmentation of the target of interest from background is a key step in image analysis. To segment vegetation pixels from background pixels, a binary mask was created by thresholding the normalized difference vegetation index (NDVI) (Rouse et al., 1973) and excessive green index (EGI) ( Moghimi, Aghkhani, Golzarian, Rohani, & Yang, 2015). The masks were then multiplied together element-wise to generate a primary mask for leaf segmentation (Figure 1 – step I). Pixels near leaf edges were likely to have spectral characteristics of mixed pixels, because they were located near the vegetation/background boundary. To assure these mixed pixels would not pass the vegetation mask, a morphological operation (erosion with 3x3 matrix of ones as structuring element) was applied on the primary mask to check the connectivity of each pixel with its neighbors. Pixels from the primary binary mask that were connected with less than eight neighbors were excluded from the final mask (Figure 1 – step II). This final mask was then used to extract all vegetation pixels from the hyperspectral data cube. The masked hyperspectral data cube was converted to a 2D matrix X whose columns were features (i.e., wavelengths) and rows were samples (i.e., pixels) and subsequent analysis was performed on matrix X. These matrices are available for each wheat line as *.mat files associated with MATLAB. References Liberatore, K. L., Dukowic-schulze, S., Miller, M. E., Chen, C., and Kianian, S. F. (2016). Free Radical Biology and Medicine The role of mitochondria in plant development and stress tolerance. Free Radic. Biol. Med. 100, 238–256. doi:10.1016/j.freeradbiomed.2016.03.033. Moghimi, A., Aghkhani, M. H., Golzarian, M. R., Rohani, A., and Yang, C. (2015). A robo-vision algorithm for automatic harvesting of green bell pepper. Am. Soc. Agric. Biol. Eng. Annu. Int. Meet. 2015 4, 2–10. doi:10.13031/aim.20152189355. Moghimi, A., Yang, C., Miller, M. E., Kianian, S., and Marchetto, P. (2017). Hyperspectral imaging to identify salt-tolerant wheat lines. Auton. Air Gr. Sens. Syst. Agric. Optim. Phenotyping II 2017. doi:10.1117/12.2262388. Munns, R., James, R. A., and Läuchli, A. (2006). Approaches to increasing the salt tolerance of wheat and other cereals. J. Exp. Bot. 57, 1025–1043. doi:10.1093/jxb/erj100. Rouse, J. W., Haas, R. H., Schell, J. A., and Deering, D. W. (1973). Monitoring vegetation systems in the Great Plains with ERTS. in Proceedings of the Third ERTS Symposium (Washington DC OR - NASA), 309–317. Available at: citeulike-article-id:9507328. Schachtman, D. P., Lagudah, E. S., and Munns, R. (1992). The expression of salt tolerance from Triticum tauschii in hexaploid wheat. Theor. Appl. Genet. 84, 714–719. doi:10.1007/BF00224174. Tsunewaki, K., Wang, G.-Z., and Matsuoka, Y. (2002). Plasmon analysis of Triticum (wheat) and Aegilops. 2. Characterization and classification of 47 plasmons based on their effects on common wheat phenotype. Genes Genet. Syst. 77, 409–27. doi:10.1266/ggs.77.409. Tsunewaki, K., Wang, G. Z., and Matsuoka, Y. (1996). Plasmon analysis of Triticum (wheat) and Aegilops. 1. Production of alloplasmic common wheats and their fertilities. Genes Genet. Syst. 71, 293–311. doi:10.1266/ggs.71.293. Zhang, Y., Liu, Z., Khan, A. A., Lin, Q., Han, Y., Mu, P., et al. (2016). Expression partitioning of homeologs and tandem duplications contribute to salt tolerance in wheat (Triticum aestivum L.). Sci. Rep. 6, 21476. doi:10.1038/srep21476. 2. Methods for processing the data: Described in response to question 1. 3. Instrument- or software-specific information needed to interpret the data: ENVI ERDAS IMAGINE Spectronon (free software available at https://downloads.resonon.com/) Programming language such as MATLAB or PYTHON 4. Standards and calibration information, if appropriate: Described in response to question 1. 5. Environmental/experimental conditions: Described in response to question 1. 6. Describe any quality-assurance procedures performed on the data: N/A 7. People involved with sample collection, processing, analysis and/or submission: N/A