We deployed object detection to find and localise thistle weeds in crop fields, drawing a tight bounding box around each individual plant — the basis for precise, plant-level weed management.
The challenge
Detecting thistle in a wheat field is genuinely hard: thistle and wheat look similar in colour and form, wheat partially hides the weeds, plants range from tiny seedlings to mature specimens, and the background mixes vegetation, soil and shadow. Treatment also needs near-real-time detection to be useful.
Our approach
We trained a state-of-the-art detector (an improved YOLO-family architecture) on high-resolution field imagery from drones, ground vehicles and fixed cameras to:
- detect thistle across the entire field of view;
- output axis-aligned bounding boxes for each plant, defined by corner coordinates (or centre, width and height).
Results & benefits
- Reliable, plant-level detection that powers spot spraying and mechanical removal.
- Inputs directed exactly where they are needed — and nowhere else.
