Abstract: --------- This dataset contains 34666 RGB-images taken from different angles and distances of weeds common in Manitoba. The imaged species common name, scientific name, and number of their images are: ______________________________________________________ | Echinochloa crus-galli | Large Barnyard Grass | 8621 | | Cirsium arvense | Canada Thistle | 4706 | | Brassica napus | Volunteer Canola | 6723 | | Taraxacum officinale | Dandelion | 4797 | | Persicaria spp. | Smartweed | 870 | | Fallopia convolvulus | Wild Buckwheat | 4165 | | Avena fatua | Wild Oat | 1218 | | Setaria pumila | Yellow Foxtail | 3566 | ------------------------------------------------------ Those single-plant images are cropped out from images originally taken by the system described in the paper available at https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0243923. The original images and their metadata information are not contained in this dataset. This dataset also contains a trained ResNet50 convolutional neural network model. It is trained to distinguish between monocots and dicots. A small collection of test datasets is included that can be used to measure the generalization capabilities of trained models. For the single-plant dataset and all test-datasets csv-files containing filenames with respective labels can be constructed by running the python-script named "parse_folders_upload.py". File Structure: --------------- Main-Directory |- readme.txt: This text file |- ResNet50_trained.hdf5: A ResNet model with 50 layers. Trained as binary classifier for dicots/monocots |- logger_binary.csv: The training results of above ResNet model |- main.py: Script used to train the above model |- parse_folders.py: Utility functions to create csv-files with filename -> label information |- apply_model.py: Script to make predictions with above model on test data |- main_data.rar: main-dataset |- test_data.rar: several test-datasets as follows... |------ Test-Data |- SameAngles: Folder containing 3494 RGB-images of plants, same format as main-dataset |- RandomAngles: Folder containing 520 RGB-images of plants, randomly chosen angles |- Smartphone: Folder containing 56 RGB-images of plants taken with smartphone camera Creators: --------- The Weedling Dataset was created at the University of Winnipeg in 2020. For contact, feedback, comments please contact Michael A. Beck (m.beck@uwinnipeg.ca), Christopher Bidinosti (c.bidinosti@uwinnipeg.ca), or Christopher Henry (ch.henry@uwinnipeg.ca). Citation, Copyright, and Use: --------- Cite the following article, when using this dataset or parts of it: @article{10.1371/journal.pone.0243923, doi = {10.1371/journal.pone.0243923}, author = {Beck, Michael A. AND Liu, Chen-Yi AND Bidinosti, Christopher P. AND Henry, Christopher J. AND Godee, Cara M. AND Ajmani, Manisha}, journal = {PLOS ONE}, publisher = {Public Library of Science}, title = {An embedded system for the automated generation of labeled plant images to enable machine learning applications in agriculture}, year = {2020}, month = {12}, volume = {15}, url = {https://doi.org/10.1371/journal.pone.0243923}, pages = {1-23} } More (Meta-)Data: ---------- The above dataset is the "minimal" dataset being used in the above article available at https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0243923. The original uncropped images and additional metadata on the images (such as camera-position and -orientation) are being planned to be made accessible. For details, please contact the creators listed above and/or follow updates on this dataset.