Invasive species detection using drones, big data and computer learning

Client
Roy Hill
Location
Western Australia

Unmanned aerial vehicles (UAV or drone) are a rapidly emerging technology revolutionising geospatial data collection. The low operational cost and fly-on-demand capabilities of UAVs, allows for the rapid and frequent collection of imagery with unprecedented detail. Through a combination of advanced image analysis techniques and machine learning, UAV data can be utilised for the fine-scale detection, classification and monitoring of invasive plant species. UAVs are an ideal tool to complement invasive species mapping and management, producing valuable fine-scale vegetation maps of large areas to efficiently direct ground efforts to maximum effect.

Roy Hill Iron Ore are proactive in the management of Parkinsonia aculeata (Parkinsonia) infestations, a Weed of National Significance posing significant environmental and economic risks. Roy Hill recently teamed up with Astron to identify and monitor Parkinsonia infestations using UAV imagery. Thematic maps of plant species distribution can be generated from imagery using image classification techniques. Imagery was collected across a trial site using Roy Hill’s Sensefly eBee UAV equipped with a near-infrared sensor. Near-infrared light is critically important for vegetation measurements, as it strongly correlates with both plant biomass and condition. Additionally, the ultra-high resolution of UAV data allows for unprecedented resolving of vegetation structural detail. However, as fine-scale components shared between vegetation become visible (e.g. woody branches), separating similar vegetation species becomes increasingly complex. To compensate for this additional complexity and to maintain classification performance, Astron employed advanced image analysis techniques, including: texture analysis, data mining and machine learning.

Variations in the growth habit of plant species often results in distinct morphological differences. Astron used image texture analysis to quantify the structure of vegetation. Given the vast selection of vegetation textural details, a machine learning approach was adopted to mine through the dataset. Using this data mining process, the most distinct structural aspects of species morphology were identified and incorporated into the image classification process.

The thematic maps generated within this pilot study revealed the distribution of Parkinsonia across an area of 173 hectares at a preliminary accuracy rate of 74%. The UAV continues to be a maturing technology, and the valuable lessons learnt through this pilot study will positively feed back into the developmental process. Improvements and experience in surveying techniques and data analysis will continue to drive increases in UAV mapping accuracy and reliability. The rapid generation of accurate vegetation maps provides invaluable support for ground teams by both identifying satellite populations and isolated individual Parkinsonia plants, as well as providing on-going monitoring of potential resurgent populations. These results clearly demonstrate the invaluable role that UAV’s when coupled with advanced image analysis techniques, will play in the ongoing management of invasive plant species.

Invasive species detection using drones, big data and computer learning
Invasive species detection using drones, big data and computer learning
Invasive species detection using drones, big data and computer learning
Invasive species detection using drones, big data and computer learning
Invasive species detection using drones, big data and computer learning
Invasive species detection using drones, big data and computer learning
Invasive species detection using drones, big data and computer learning : Image 1
Invasive species detection using drones, big data and computer learning : Image 2
Invasive species detection using drones, big data and computer learning : Image 3
Invasive species detection using drones, big data and computer learning : Image 4
Invasive species detection using drones, big data and computer learning : Image 5
Invasive species detection using drones, big data and computer learning : Image 6