Community

OpenWeedLocator is an open-source project built by researchers and farmers. Whether you’re building your first unit, training weed detection models, or improving the codebase, there are several ways to get involved.

Contributing

We welcome contributions of all kinds:

  • Bug reports and feature requests — open an issue on GitHub

  • Code contributions — fork the repository, make your changes, and submit a pull request

  • Weed detection models — share trained models or annotated datasets with the community

  • Field testing — report your experiences with different crops, weeds, and conditions

  • Documentation — help improve these docs or translate them

See the Development Setup guide for getting started with the codebase on your desktop.

Publications

If you use OWL in your research, please cite the relevant publications below.

OWL platform and fallow weed detection

The original paper introducing the OpenWeedLocator platform, its design, and validation across seven fallow fields in New South Wales, Australia.

Coleman, G., Salter, W. and Walsh, M. (2022). OpenWeedLocator (OWL): an open-source, low-cost device for fallow weed detection. Scientific Reports, 12(1), 170. doi:10.1038/s41598-021-03858-9

Key findings:

  • Four colour-based detection algorithms (ExG, ExGR, ExHSV, HSV) were validated for fallow weed detection

  • Average precision of 79% and recall of 52% across all fields and algorithms

  • Individual transects achieved up to 92% precision and 74% recall

  • Total hardware cost under AU$400, making site-specific weed control accessible to smaller operations

  • A comprehensive GitHub repository was developed to promote community-driven technology development in agriculture

BibTeX

```bibtex @article{Coleman2022, author = {Coleman, Guy and Salter, William and Walsh, Michael}, title = , journal = {Scientific Reports}, volume = {12}, number = {1}, pages = {170}, year = {2022}, doi = {10.1038/s41598-021-03858-9} } ```

Speed and camera performance

A follow-up study investigating the effect of ground speed (5–30 km/h) and camera hardware on weed detection performance, using tillage radish (Raphanus sativus) and forage oats (Avena sativa) as representative broadleaf and grass weeds.

Coleman, G.R.Y., Macintyre, A., Walsh, M.J. and Salter, W.T. (2023). Investigating image-based fallow weed detection performance on Raphanus sativus and Avena sativa at speeds up to 30 km h^-1^. Computers and Electronics in Agriculture, 215, 108419. doi:10.1016/j.compag.2023.108419

Key findings:

  • Four camera/software combinations were tested on the OWL platform: Raspberry Pi HQ camera (default and optimised settings), Raspberry Pi v2 camera, and an Arducam AR0234 global shutter camera

  • The global shutter Arducam AR0234 achieved the highest recall — up to 100% for broadleaf weeds and 91.6% for grass weeds at 5 km/h

  • At 30 km/h, the Arducam’s grass weed recall declined by 12.4%, but broadleaf recall was not significantly affected

  • All cameras experienced decreasing recall with increasing speed — the default HQ camera showed the steepest decline at 1.12% per km/h for broadleaf weeds

  • Detection of grass weeds (forage oats) was significantly worse than broadleaf weeds (tillage radish) across all cameras

  • Despite variations in recall, the HQ and v2 cameras maintained near-perfect precision at all tested speeds

  • Results highlight the importance of camera selection and software tuning for real-world deployment

BibTeX

```bibtex @article{Coleman2023, author = {Coleman, Guy R.Y. and Macintyre, Angus and Walsh, Michael J. and Salter, William T.}, title = , journal = {Computers and Electronics in Agriculture}, volume = {215}, pages = {108419}, year = {2023}, doi = {10.1016/j.compag.2023.108419} } ```

Datasets

Annotated weed image datasets are available on Weed-AI, a community platform for sharing weed detection training data. If you collect and annotate images for your own OWL deployment, consider contributing them back.

Getting Help

Licence

OpenWeedLocator is released under the MIT Licence.