Leaf venation networks of Bornean trees: images and hand-traced segmentations
The dataset contains images and tracings of leaf venation networks obtained from tree species in Malaysian Borneo. These images are suitable for biological analysis and/or for training of machine-learning algorithms. Samples were obtained from eight permanent forest plots that are part of the Globa...
| Main Author: | |
|---|---|
| Format: | Dataset |
| Published: |
University of Oxford
2018
|
| Subjects: |
| _version_ | 1797106537351610368 |
|---|---|
| author | Blonder, B |
| author2 | Blonder, B |
| author_facet | Blonder, B Blonder, B |
| author_sort | Blonder, B |
| collection | OXFORD |
| description | The dataset contains images and tracings of leaf venation networks obtained from tree species in Malaysian Borneo. These images are suitable for biological analysis and/or for training of machine-learning algorithms.
Samples were obtained from eight permanent forest plots that are part of the Global Ecosystems Monitoring network (http://gem.tropicalforests.ox.ac.uk) and managed through the Biodiversity And Land-use Impacts on tropical ecosystem function (BALI) project (http://bali.hmtf.info). The plots are selected to span a logging intensity gradient and are characterized by a mixed dipterocarp lowland forest in various stages of regeneration. Each plot is 1 hectare in size, with all stems > 10 cm diameter at breast height tagged.
Leaves were sampled from the dominant species comprising 80% of the total basal area in each plot. A single sunlit and/or shaded leaf from each stem was sampled using single rope climbing techniques. This leaf was cleaned using a wet rag and then pressed flat and dried at 60°C for at least one week.
1 cm^2 sections were cut from each leaf, selecting portions in the middle of the lamina that avoided any primary veins, and chemically cleared following standard protocols. Briefly, leaf samples were immersed in warm 5% aqueous sodium hydroxide until transparent (3-7 days), rinsed in water, and then bleached in 2.5% aqueous sodium hypochlorite for 5 min. After another water rinse, samples were passed through an ethanol dehydration series to 100% ethanol, then lignin stained in 0.1% safranin in ethanol. After a 100% ethanol rinse, samples were passed through a dilution series to 100% toluene and then mounted on a glass slide in a toluene-based mounting medium (Richard-Allen Scientific).
After allowing slides to dry for 3 days, each leaf was imaged using a compound microscope (Olympus, BX43) with 2x apochromat objective and an Olympus SC100 color camera (3840 x 2748 pixel resolution). 9-16 overlapping image fields were stitched together for each sample to obtain a complete image of the sampled area using GIMP software (version 2.8, https://www.gimp.org/).
Images were pre-processed by retaining the green channel of each image which has the greatest contrast after safranin staining, and applying contrast-limited adaptive histogram equalization, with a 400×400 pixel window and a contrast limit of 0.01.
A polygonal region-of-interest (ROI) around 700×700 pixels was manually traced for each image to exclude background and damaged pixels, using a digitizing tablet (Wacom, Cintiq 22HD, http://www.wacom.com/) and GIMP software. All the veins within this region were traced, with the traced vein widths corresponding to the apparent widths in images. In addition, any substantive veins with width greater than ~0.5 mm were also manually delineated through the entire image, as these were not routinely included in tracings.
Each sample is represented by a CODE with format X-TY-BZ. X indicates the name of a plot in the Global Ecosystems Monitoring network database (e.g. 'BEL'), Y indicates the number of a tree within a plot (e.g. '101') and Z represents the light stratum of the canopy where the leaf was collected (either 'S' for 'sunlit' or 'SH' for 'shaded'). A small number of samples have format X-GY-BZ, where GY indicates a plant (number Y) sampled from a sunlit gap.
Each sample is then represented by 4-5 image files,
CODE_img.png: the raw grayscale image.
CODE_roi.png: a binary mask indicating the hand-traced region of interest
CODE_seg.png: a binary mask indicating the hand-traced presence/absence of veins within the region of interest
CODE_slc.png: a binary mask indicating the hand-traced presence/absence of leaf tissue within the image (of high image quality, i.e. excluding tears, bubbles, dust)
CODE_big.png: a binary mask indicating the hand-traced presence/absence of larger veins within the region of interest.
The first 4 files are always present; the last (_big) only when it was determined that larger veins were present.
A metadata file linking CODE to taxonomy is also present (Metadata.csv). Each CODE is associated with a family, genus, and species.
The final dataset contains 726 leaves from 295 species comprising 50 families. All images have a resolution of 595 pixels per millimeter. Image extents are approximately 8000 x 8000 pixels. |
| first_indexed | 2024-03-07T07:03:53Z |
| format | Dataset |
| id | oxford-uuid:de65fc07-4b8f-4277-a6c4-82836afbdeb3 |
| institution | University of Oxford |
| last_indexed | 2024-03-07T07:03:53Z |
| publishDate | 2018 |
| publisher | University of Oxford |
| record_format | dspace |
| spelling | oxford-uuid:de65fc07-4b8f-4277-a6c4-82836afbdeb32022-03-31T17:12:26ZLeaf venation networks of Bornean trees: images and hand-traced segmentationsDatasethttp://purl.org/coar/resource_type/c_ddb1uuid:de65fc07-4b8f-4277-a6c4-82836afbdeb3image analysisecologybotanymachine learningplant ecophysiologytropical forestsORA DepositUniversity of Oxford2018Blonder, BBlonder, BBoth, SJodra, MMajalap, NBurslem, DTeh, YMalhi, YThe dataset contains images and tracings of leaf venation networks obtained from tree species in Malaysian Borneo. These images are suitable for biological analysis and/or for training of machine-learning algorithms. Samples were obtained from eight permanent forest plots that are part of the Global Ecosystems Monitoring network (http://gem.tropicalforests.ox.ac.uk) and managed through the Biodiversity And Land-use Impacts on tropical ecosystem function (BALI) project (http://bali.hmtf.info). The plots are selected to span a logging intensity gradient and are characterized by a mixed dipterocarp lowland forest in various stages of regeneration. Each plot is 1 hectare in size, with all stems > 10 cm diameter at breast height tagged. Leaves were sampled from the dominant species comprising 80% of the total basal area in each plot. A single sunlit and/or shaded leaf from each stem was sampled using single rope climbing techniques. This leaf was cleaned using a wet rag and then pressed flat and dried at 60°C for at least one week. 1 cm^2 sections were cut from each leaf, selecting portions in the middle of the lamina that avoided any primary veins, and chemically cleared following standard protocols. Briefly, leaf samples were immersed in warm 5% aqueous sodium hydroxide until transparent (3-7 days), rinsed in water, and then bleached in 2.5% aqueous sodium hypochlorite for 5 min. After another water rinse, samples were passed through an ethanol dehydration series to 100% ethanol, then lignin stained in 0.1% safranin in ethanol. After a 100% ethanol rinse, samples were passed through a dilution series to 100% toluene and then mounted on a glass slide in a toluene-based mounting medium (Richard-Allen Scientific). After allowing slides to dry for 3 days, each leaf was imaged using a compound microscope (Olympus, BX43) with 2x apochromat objective and an Olympus SC100 color camera (3840 x 2748 pixel resolution). 9-16 overlapping image fields were stitched together for each sample to obtain a complete image of the sampled area using GIMP software (version 2.8, https://www.gimp.org/). Images were pre-processed by retaining the green channel of each image which has the greatest contrast after safranin staining, and applying contrast-limited adaptive histogram equalization, with a 400×400 pixel window and a contrast limit of 0.01. A polygonal region-of-interest (ROI) around 700×700 pixels was manually traced for each image to exclude background and damaged pixels, using a digitizing tablet (Wacom, Cintiq 22HD, http://www.wacom.com/) and GIMP software. All the veins within this region were traced, with the traced vein widths corresponding to the apparent widths in images. In addition, any substantive veins with width greater than ~0.5 mm were also manually delineated through the entire image, as these were not routinely included in tracings. Each sample is represented by a CODE with format X-TY-BZ. X indicates the name of a plot in the Global Ecosystems Monitoring network database (e.g. 'BEL'), Y indicates the number of a tree within a plot (e.g. '101') and Z represents the light stratum of the canopy where the leaf was collected (either 'S' for 'sunlit' or 'SH' for 'shaded'). A small number of samples have format X-GY-BZ, where GY indicates a plant (number Y) sampled from a sunlit gap. Each sample is then represented by 4-5 image files, CODE_img.png: the raw grayscale image. CODE_roi.png: a binary mask indicating the hand-traced region of interest CODE_seg.png: a binary mask indicating the hand-traced presence/absence of veins within the region of interest CODE_slc.png: a binary mask indicating the hand-traced presence/absence of leaf tissue within the image (of high image quality, i.e. excluding tears, bubbles, dust) CODE_big.png: a binary mask indicating the hand-traced presence/absence of larger veins within the region of interest. The first 4 files are always present; the last (_big) only when it was determined that larger veins were present. A metadata file linking CODE to taxonomy is also present (Metadata.csv). Each CODE is associated with a family, genus, and species. The final dataset contains 726 leaves from 295 species comprising 50 families. All images have a resolution of 595 pixels per millimeter. Image extents are approximately 8000 x 8000 pixels. |
| spellingShingle | image analysis ecology botany machine learning plant ecophysiology tropical forests Blonder, B Leaf venation networks of Bornean trees: images and hand-traced segmentations |
| title | Leaf venation networks of Bornean trees: images and hand-traced segmentations |
| title_full | Leaf venation networks of Bornean trees: images and hand-traced segmentations |
| title_fullStr | Leaf venation networks of Bornean trees: images and hand-traced segmentations |
| title_full_unstemmed | Leaf venation networks of Bornean trees: images and hand-traced segmentations |
| title_short | Leaf venation networks of Bornean trees: images and hand-traced segmentations |
| title_sort | leaf venation networks of bornean trees images and hand traced segmentations |
| topic | image analysis ecology botany machine learning plant ecophysiology tropical forests |
| work_keys_str_mv | AT blonderb leafvenationnetworksofborneantreesimagesandhandtracedsegmentations |