zenodoopen

Irish Grass Clover Dataset (VistaMilk)

Irish grass clover dataset Overview The Irish Grass Clover Dataset, collected in Ireland by a group of researchers working at the  VistaMilk Science Foundation Ireland Research Centre , comprises high-resolution images of herbage biomass. The dataset includes ground-truth annotations for herbage mass (kg DM/ha) and post-cutting herbage height (cm), capturing the composition percentages of grass, weed, and clover. This document provides details about the dataset, including collection methods, image specifications, and associated ground-truth data. Dataset description The dataset contains multiple subsets of data collected using a tripod mounted camera, a handheld phone or a drone. The dataset can be divided into two main subsets: the Camera & Phone Images subset and the Drone Images subset.  Camera & Phone Images: This dataset was gathered in 2020 in Ireland using both a high-resolution Canon camera (Canon EOS 90D Camera - Canon Europe) and a smartphone camera (iPhone 6). All the images were collected at Moorepark Farm managed by Teagasc, aiming to capture the biomass composition comprising grass, weeds and clover. Each image corresponded to a 0.5×0.5 m quadrat with 5-6 images taken for each of the 26 plots. Ground truth (GT) is provided for a subset of these images while other images are collected without GT. The herbage within quadrats for which GT was collected was harvested at 2-4 cm above ground level using Gardena hand shears (Accu 60, Gardena International GmbH, Ulm, Germany) immediately after image capture. Fresh weight was recorded and the harvested herbage was separated, oven-dried for 16 hours and weighed to give dry matter yield. For each of the  GT images the following labels are provided: total dry herbage mass (kg DM/ha), dry grass biomass percentage (%), dry clover biomass percentage (%), dry weed biomass percentage (%), fresh grass biomass percentage (%), fresh clover biomass percentage (%), fresh weed biomass percentage (%), and sward height post-cutting (cm). Additional views were taken using the smartphone for validation and generalisation purposes. All labelled phone images were collected at the exact same quadrats/locations where some of the camera images were taken, therefore they share the same GT values. A larger number of unlabeled camera & phone images were also collected at random locations across the same plots where the labelled images were collected. The contents of this subset can be summarised as: * 525 GT camera images divided into 418 train set labelled images and 107  validation set labelled images. * 124 GT phone  images divided into 17 train set labelled images and 107  validation set labelled images. * 1072 unlabelled camera images  * 1112 unlabelled phone images    Drone Images An extension of the Camera & Phone Images subset was created in late Autumn of 2021 where drone images were collected in the same 23 herbage paddocks originally studied with camera & phone images. At each paddock between 7 to 36 drone images at an altitude between 6 and 12 metres were captured. The drone used is the DJI Mavic 2 Pro 1 with its default camera, taking pictures at a resolution of 5472 × 3648. A total of 331 drone images with their associated altitude were obtained. Because of the huge areas covered by drone images, the ground-truth we collect is limited to the dry herbage mass at the paddock level and we omit the grass height and biomass percentage information. Two ground-truth estimation methods were utilised for the drone images: the first is a visual estimation performed on site at the time of the image collection by two human experts, the second is following the protocol of Egan et al. [1], where two 1.2 × 8 metres strips in the paddocks are cut at 4 cm above ground level (typical cow grazing height) using an Etesia lawn mower (Etesia UK. Ltd., Warwick, UK). A 100 grams sample is collected from the cut material and dried at 95°C for 16 hours to obtain the dry herbage mass. Structure The file structure for this repository is as follows: irish_dataset_all/         |_README.md         |         |_camera         |        |_train.csv // training images (filenames and GT annotations)         |        |_train_red.csv // a smaller training subset of 52 images     names and annotations         |        |_val.csv // validation images names and annotations         |        |_images/              // The training and test camera images         |_phone         |        |_phone_gt_train.csv // training images (filenames and GT annotations)         |        |_phone_gt_train.csv // validation images (filenames and annotations)         |        |_images/              // training and test phone images         |                 |_camera_unlab                     // unlabeled Camera images         |         |_phone_unlab                     // unlabeled Phone images         |         |_drone                                      |        |_images             // Drone images, organized by paddock         |        |_labels.csv             // image filenames and associated paddock level herbage mass ground truth (including visually estimated)         |        |_paddock_gt.csv     // paddock level herbage mass ground truth (including visually estimated) License: This dataset is provided under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).  This means you are free to: Share: Copy and redistribute the material in any medium or format. Adapt: Remix, transform, and build upon the material. However, these permissions are subject to the following terms: Attribution: You must give appropriate credit, provide a link to the license, and indicate if changes were made. Non-Commercial: You may not use the material for commercial purposes. ShareAlike: If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original. By using this dataset, you agree to abide by these terms. For more details about the license, visit https://creativecommons.org/share-your-work/cclicenses/. Acknowledgements and fair use: Please cite us if our work and data helps your research ! @inproceedings{albert2021semi,   title={Semi-supervised dry herbage mass estimation using automatic data and synthetic images},   author={Albert, Paul and Saadeldin, Mohamed and Narayanan, Badri and Mac Namee, Brian and Hennessy, Deirdre and O'Connor, Aisling and O'Connor, Noel and McGuinness, Kevin},   booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},   pages={1284--1293},   year={2021} } @inproceedings{albert2022unsupervised,   title={Unsupervised domain adaptation and super resolution on drone images for autonomous dry herbage biomass estimation},   author={Albert, Paul and Saadeldin, Mohamed and Narayanan, Badri and Mac Namee, Brian and Hennessy, Deirdre and O'Connor, Noel E and McGuinness, Kevin},   booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},   pages={1636--1646},   year={2022} } @article{albert2022utilizing,   title={Utilizing unsupervised learning to improve sward content prediction and herbage mass estimation},   author={Albert, Paul and Saadeldin, Mohamed and Narayanan, Badri and Mac Namee, Brian and Hennessy, Deirdre and O'Connor, Aisling H and O'Connor, Noel E and McGuinness, Kevin},   journal={arXiv preprint arXiv:2204.09343},   year={2022} } References [1] Egan, Michael, Norann Galvin, and Deirdre Hennessy. "Incorporating white clover (Trifolium repens L.) into perennial ryegrass (Lolium perenne L.) swards receiving varying levels of nitrogen fertilizer: Effects on milk and herbage production." Journal of Dairy Science 101, no. 4 (2018): 3412-3427 [a]Get license added here.

ShareScore

28/100

Overall dataset sharing score

Score breakdown

These five areas show where the dataset supports — or may limit — practical reuse.

Stewardship: 8
Harmonization: 4
Access: 16
Reuse readiness: 0
Engagement: 0