CoFly-WeedDB dataset (~436MB) consists of 201 aerial images, capturing different weed types that interfere among the row crop (cotton), as well as their corresponding annotated images. The annotation process of the weed instances was conducted by agronomists, using the LabelMe annotation tool, indicating three different types of weeds:
- Johnson grass (Sorghum halepence)
- Field bindweed (Convolvulus arvensis)
- Purslane (Portulaca oleracea)
Regarding the annotated images, each weed type is notated with a different color while the rest of the image is considered as background. In specific:
- Red-Johnson grass
- Yellow-Field bindweed
- Blue-Purslane
- Black-Background
CoFly-WeedDB dataset can be downloaded here.
This dataset was created using an RGB camera (1-inch 20-megapixel CMOS sensor) mounted on a DJI Phantom 4 Pro UAV. The RGB images were collected while the UAV was performing a coverage mission over the field's area. During the designed mission, the camera angle was adjusted to -87°, vertically with the field. The flight altitude and speed of the UAV were equal to 5m and 3m/s respectively, aiming to provide a close and clear view of the weed instances.
Examples of the CoFly-WeedDB dataset.
 |
 |
 |
 |
 |
 |
| (a) RGB | (b) Annotated mask | (c) Overlay |
Figure 1. Samples of the CoFly-WeedDB dataset: (a) RGB image, (b) annotated mask and (c) overlay. Red color corresponds to johnson grass, yellow to field bindweed, blue to purslane and black is for background.
- 'images' folder: contains the 1280x720 RGB images. Total images are 366 but 201 are annotated since the rest do not enclose any weed instances.
- 'labels' folder: contains the corresponding annotated masks of 201 RGB images.
- 'labels_1d' folder: contains the ground truth masks in a classification format. Pixel values correspond to classes, 0: background, 1: johnson grass, 2: purslane, 3: field bindweed.
- 'overlay' folder: contains an overaly of original RGB images and corresponding annotation masks.
Image filenames are defined with an ID number, such as:
ID_00048.png
A code snippet titled mask2box.py is provided to convert the segmentation masks to bounding boxes for object detection oriented tasks.
If you use this dataset, please cite the following publication:
Krestenitis, M., Raptis, E. K., Kapoutsis, A. C., Ioannidis, K., Kosmatopoulos, E. B., Vrochidis, S., & Kompatsiaris, I. (2022). CoFly-WeedDB: A UAV image dataset for weed detection and species identification. Data in Brief, 45, 108575. [Link]
@article{krestenitis2022cofly,
title={CoFly-WeedDB: A UAV image dataset for weed detection and species identification},
author={Krestenitis, Marios and Raptis, Emmanuel K and Kapoutsis, Athanasios Ch and Ioannidis, Konstantinos and Kosmatopoulos, Elias B and Vrochidis, Stefanos and Kompatsiaris, Ioannis},
journal={Data in Brief},
volume={45},
pages={108575},
year={2022},
issn = {2352-3409},
doi = {https://doi.org/10.1016/j.dib.2022.108575},
publisher={Elsevier}
}This research has been financed by the European Regional Development Fund of the European Union and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH - CREATE - INNOVATE (T1EDK-00636).