Understanding where and what species of vegetation are present is essential for managing wetlands. Using drones in combination with machine learning to map vegetation is one technique, which can provide more coverage than field surveys alone. Machine learning models can be taught to recognise the different colours (called spectra) of vegetation and therefore identify species.
This trial developed a model which was capable of identifying vegetation species with an overall accuracy of over 90%.
The study also investigated whether aerial imagery could be used instead of drones and found that the overall accuracy would be 85% if aerial imagery was used.
While this trial was successful in mapping the vegetation, if applying this method to other wetlands or at different times of the year, the dynamic nature of wetland vegetation (greening and drying) needs to be carefully considered and accounted for.
The code developed in this project is available at GitHub - MDBAuth/Drone-vegetation-mapping-Gwydir
The drone imagery is available upon request from DataServices@mdba.gov.au
| Publication title | ||||
|---|---|---|---|---|
| Publication title | Published | File type | File size | |
| Mapping of non-woody vegetation in sites within the Gwydir Wetlands using multispectral sensors onboard unmanned aerial platforms | 28 Oct 2024 |
PDF
|
2.99 MB | |
Published date: 28 October 2024