Training AI to hunt for weeds
A consortium led by North Coast Local Land Services, in collaboration with Charles Sturt University and Southern Cross University, is leveraging Artificial Intelligence (AI) technology to combat the invasive plant Tropical Soda Apple (TSA) in northern NSW.
The project, achieving significant milestones, involves training an AI model on drone imagery, demonstrating an impressive overall accuracy of 96-97%, offering a promising and cost-effective solution for TSA detection in rugged grazing landscapes.
In a novel initiative to combat the invasive plant Tropical Soda Apple (TSA) in northern NSW, a consortium has been diligently working on the development of remote detection technology for TSA by training an AI model. The project - funded through a grant from the Commonwealth Government - aims to enhance TSA control in rugged high-value grazing country and has already achieved significant milestones.
The consortium, coordinated by North Coast Local Land Services (LLS) and consisting of Charles Sturt University (CSU), Southern Cross University (SCU), and relevant Local Council staff, have successfully identified sites across the region, contacted and engaged eligible landholders and undertaken a pilot study.
The study has laid the groundwork for future success. Establishing study plots, estimating TSA cover abundance, and conducting drone surveys were integral components. Notably, the drone detection model results revealed an outstanding overall accuracy of 96-97%, emphasising the effectiveness of the AI model trained on drone imagery. The satellite imagery is currently being analysed and if the AI model proves accurate this could drastically decrease cost, increase accessibility and the models’ capabilities.
Drone surveys in the Bellingen Local Government Area have just been completed covering more challenging terrain where TSA is interspersed with similar looking species and obscured by long grass. This increased difficulty will allow the AI model to be become familiar with the target species in more complex environments.
The final phase of the project will include an online webinar and on-farm demonstrations to showcase results. These efforts will inform development of TSA management recommendations, providing a holistic approach to combating this invasive species. This project exemplifies the power of technology, collaboration, and innovation in addressing environmental challenges. Advances made in drone and AI technology promise a more effective and sustainable approach to weed control in agricultural landscapes.