CQU researcher gives invasive weeds the YOLO treatment
By Greg Chapman
The fight against the invasive plant Lantana camara could receive a significant boost with a CQUniversity researcher’s project using deep learning technology to help farmers identify the species early.
The project by CQU academic and AI specialist Wie Kiang Hi for his Master of Research thesis in 2023, centred around developing an intelligent system designed to identify and manage the presence of the weed by creating a catalogue of images and data sourced from existing repositories.
“The methodology employed in my study leveraged cutting-edge technologies, specifically deep learning algorithms, to train a computer model to recognise the distinct features of Lantana camara,” Wie said.
“This is akin to teaching the computer to 'see' and distinguish this invasive plant from other vegetation by analysing images.”
Wie said the research has the potential to prevent the establishment and spread of Lantana camara, and reduce the need for costly eradication measures and minimise potential crop damage.
“I obtained substantial data from third-party sources. The data collection process involved sourcing diverse image datasets of Lantana camara from reliable and established repositories and collaborating institutions,” he said.
“By leveraging data from various third-party sources, I aimed to ensure the representation of different geographical locations and ecological conditions in the dataset. This approach enhances the robustness and generalisability of the deep learning model, allowing it to identify Lantana camara effectively in a range of real-world scenarios.”
A key component of his methodology was the utilisation of a notable focus on the YOLO (You Only Look Once) algorithm.
YOLO is a widely recognised and robust computer vision algorithm that excels in object detection tasks. Its unique 'one-shot' approach allows for real-time processing, making it particularly well-suited for identifying Lantana camara swiftly and accurately within images.
Wie said although the research involved data analysis and model development, there was the potential for fieldwork or on-site activities on farms in the future.
“My research endeavoured to offer practical solutions for farmers in the ongoing battle against invasive plants. By providing an advanced tool for early detection and management, I hope to contribute to sustainable agriculture, preserve native ecosystems, and enhance the resilience of farming communities.”
