Blogs

Weed control has always been part of the Landcare ethic. In the Desert of Queensland, Prickly Acacias cover around 19 million of hectares. They are a threat to the Australian environment as they kill the pasture needed to feed the livestock. The economic impact on Queensland’s grazing industry was estimated at $5 million per year in 2003 when they were only covering 6.6 million of hectares.

For 10 years, DCQ has been aiming at eradicate them in order to restore the native grass. As Prickly Acacias are widely spread, DCQ and Sunbirds partnered in 2019 and 2020 to test the use of long-range fixed wing drones to monitor the weed.

Objectives

In June 2019, a first mission has been set up to map 800 hectares of Prickly Acacias in the region of Longreach. The objectives were:

• to demonstrate the viability and the added-value of using drones to cover large areas compared to helicopters and trucks
• to collect aerial data of Prickly Acacias: high resolution photos and geolocalisation
• to test if the data collected are precise and good enough to be used in the artificial intelligence (AI) algorithms which identify and classify the trees.

Method/approach

To train the AI model to automatically detect Prickly Acacias from photos, a significant quantity of pictures needed to be taken by the drones first. The method was the following:

1. Identify the zones of interest with a certain density of Prickly Acacias
2. Identify the time of the year to survey: Prickly Acacias flower between March and June, time of the year where they are the most visible.
3. Identify the time of the day to survey: in order to have a great sunlight in the pictures, flights were only done from 10am to 3pm, and not during cloudy days in order to have a consistent light on the pictures.
4. Fly the drone on the designated area: 10 days have been necessary to obtain data on 800 hectares.
5. Use the pictures taken by the drones to create a single high definition map of the areas.
6. Use the high definition map to train DCQ AI model to recognize Prickly Acacias

Key findings

1. Outback proven drones: they were able to fly as normal with minimal impact from the dust and the windy conditions
2. Hundreds of geolocalized photos have been generated
3. With a resolution of 2.78 centimetres, the quality and precision of the pictures were excellent and highly useable to run the AI model
4. The aerial photos have then been used to select and classify trees by size in the model.
5. Using the data, DCQ have been able to reduce the quantity of pesticide needed for their eradication program, reducing operational costs.

Conclusions

DCQ became the premier weed control organisation across one-third of Queensland. These types of solutions enable communities to better address long term strategic issues of national importance.