AI-Powered Detection of Chilean Needle Grass for Mine Recultivation

Earlier this year, we announced that Proofminder had been selected for the Eureka Innowwide, funded by the European Union, to develop and deploy an AI-powered weed detection system in one of the toughest environments in Australia — mine site rehabilitation.  

This project targets one of the country’s most damaging invasive weeds, Chilean needle grass (CNG) — a declared Weed of National Significance (WoNS) that undermines agriculture, biodiversity, and large-scale land restoration efforts. 

The Challenge: A Persistent and Costly Threat

CNG is no ordinary weed. 

  • It already affects 40 million hectares of Australia and has the potential to spread to 180 million hectares, or 25% of the country’s landmass. 
  • It can reduce productivity by up to 50% and outcompete native vegetation, forming dense monocultures that halt biodiversity recovery. 
  • Each plant can produce 10,000–20,000 seeds per year, many of which remain viable in the soil for over a decade. 

For mine rehabilitation projects, this weed is more than an ecological issue — it’s a compliance and financial risk. 

  • Under the NSW Biosecurity Act 2015, mining leaseholders must prevent and control WoNS weeds. 
  • CNG infestations can delay closure certification, trigger repeated rework and reseeding, and add $2,000–$5,000+ per hectare per year in rehabilitation costs. 

Detection is equally challenging: 

  • Rehabilitation areas can be vast and remote. 
  • Manual inspections are slow, labour-intensive, and prone to missing early-stage plants. 
  • Regulatory sign-off requires consistent proof of control over several years. 

The Solution: High-Resolution Aerial Imaging + AI Detection

Through Eureka Innowwide funding, Proofminder has developed a specialised image capture and AI analysis process tailored for mine site conditions. 

We began by identifying the optimal flight parameters for image gathering — from altitude and speed to camera settings — to ensure our dataset met the precision requirements for training. 

Using a DJI Enterprise drone, we collect imagery at a resolution of just a few millimetres per pixel. This ultra-high detail allows our AI models to detect CNG plants at early growth stages, even when they are visually similar to surrounding vegetation. 

Once the images are captured: 

  • The AI analyses the dataset to locate individual plants with millimetre-level accuracy. 
  • Each detection is logged with exact GPS coordinates. 
  • These coordinates are exportable for any type of treatment — tractor-mounted booms, robotic sprayers, or manual spot application. 

Results to Date: From Field Data to Trained Model 

Since the project began, we have: 

  1. Defined and validated flight parameters that deliver consistent image quality across varied mine site environments. 
  2. Captured and annotated extensive training datasets at a few mm/px resolution. 
  3. Trained and tested the AI model on this data, achieving high detection accuracy across different light and soil conditions. 
  4. Produced precision weed maps fully compatible with existing spraying machinery, robotics, and manual treatment workflows. 
  5. Created regulator-ready reporting outputs that provide verifiable records of weed presence or absence. 

The model is now successfully trained and field-tested, ready for scaling to larger rehabilitation areas. 

Impact Potential

By using this approach, mine operators can: 

✅ Accelerate rehabilitation by removing CNG before it spreads and dominates.

✅ Reduce chemical and labour costsby $2,000–$5,000 per hectare through targeted treatment.

✅ Restore native vegetation and meet biodiversity targets faster.

✅ Prevent off-site spread, protecting surrounding farmland and ecosystems.

✅ Maintain compliance with biosecurity and rehabilitation obligations.

Next Steps 

The next stage of the project will focus on: 

  • Expanding AI model training with seasonal and environmental variability. 
  • Demonstrating consistent performance at scale across multiple mine sites. 
  • Quantifying cost savings and environmental benefits over consecutive monitoring cycles. 

Chilean needle grass is one of Australia’s most persistent and costly invasive weeds — but with the right flight parameters, high-resolution aerial imagery, and AI-powered detection, it can be located early, targeted precisely, and controlled effectively.

Thanks to Eureka Innowwide’s EU funding, this project is demonstrating that even in the most challenging rehabilitation environments, advanced imaging and AI can deliver measurable environmental, economic, and compliance benefits.

👉 Interested in learning more or booking a demo? Contact us via the form below.  

Lets talk about your project

Submit the form to get started