Effective weed management is critical to improving agricultural productivity, reducing unnecessary chemical use, and supporting more sustainable farming practices. Current weed detection and mapping approaches can be limited by variability in weed species, crop conditions, growth stages and field environments, and false detections can reduce confidence in automated systems. Advances in artificial intelligence, computer vision and remote sensing provide an opportunity to develop more reliable and adaptable tools that can identify, locate and characterise weeds under real-world agricultural conditions.
This project will build on emerging AI approaches for visual understanding and object localisation, including methods capable of distinguishing target weeds from crops and other non-target objects. Importantly, the technology will be designed to recognise when a target is not present rather than forcing a potentially incorrect detection, supporting more reliable weed mapping and targeted management.
Aim: The project aims to develop and evaluate an AI-enabled framework for improved weed detection, mapping and management in agricultural environments. The overarching goal is to generate reliable, spatially relevant information that can support better decisions about where, when and how weeds should be managed.
Specific activities will investigate adaptable AI methods for detecting and differentiating weeds from crops and other vegetation, reducing false detections, and translating observations into information suitable for practical weed-management decisions. The resulting capability could support applications including weed mapping, targeted or spot spraying, variable-rate treatment, field monitoring and assessment of weed-management outcomes.
Meet the PhD student on this project
Xin Zhu
My name is Xin Zhu. I hold a Master’s degree from USTC and a Bachelor’s degree from SWPU, both in Software Engineering. I spent 4 years in academia (Unisa, CUHK) as a Research Assistant, and I have over 10 years of experience in the IT industry (Huawei, NeuSoft, a Startup, Sensetime). My past work focused on the development of ocean storage systems (io multipath, backup, and recovery), causal inference, deep-learning-based computer vision (ImageNet 2016, object perception, crack seg, …), robotics (arm and vacuum cleaner), and autonomous driving systems (control, sensor calibration, mapping, data mining).
I am currently developing an innovative, deployment-ready deep learning framework for accurate weed detection and segmentation from aerial imagery of sugarcane fields.
Outside of research, I enjoy playing soccer, walking, and hiking.
Supervisors and advisors
Mostafa Rahimi Azghadi, Mohammad Jahanbakht and Tao Huang, James Cook University
Approach
The project will adopt a flexible, data-driven approach combining AI, computer vision and spatial agricultural data. Depending on the application and available datasets, imagery may be collected from ground-based cameras, agricultural machinery, drones or other sensing platforms. AI models will be developed and evaluated for identifying and localising weeds, distinguishing relevant weed characteristics, and reliably determining when target weeds are absent. The underlying approach builds on the concept demonstrated in the attached work, where a generalised model can be adapted to agricultural applications with relatively limited domain-specific fine-tuning.
The project will also investigate how AI-derived weed information can be converted into actionable weed-management outputs, such as weed distribution and density maps, priority treatment areas and information to support targeted spraying or other management interventions. Field validation and engagement with end users will be used where appropriate to assess accuracy, robustness and practical value, while maintaining sufficient flexibility to incorporate different crops, weed species, sensing technologies and management requirements as the project develops.
Want to know more? Email Sarah Adams, Chief Operating Officer, at plantbiosecurity@anu.edu.au or +61 459 341 281.