Weed detection good practices - PIX4Dfields
Weed detection ranks among the most widely adopted drone applications in agriculture. The optimal imagery and resolution depends on the specific weed species present and the timing within the growing season.
Fallow weed detection
When weeds are detected in a fallow field, vegetation indices are generally used to differentiate the green weeds from the bare soil or previous crop residue. These indices can be derived from both RGB and multispectral cameras.
RGB vs. Multispectral in fallow weed detection
At the same flight altitude, multispectral images, even with lower resolution, help to better differentiate weeds.
- Multispectral images are superior to RGB for detecting green weeds in fallow fields, especially when using the NDVI and OSAVI indices. The bad part is that
Weeds with a 10 cm diameter can be detected in fallow fields using 4cm GSD MS. - RGB-based methods (VARI, TGI) are less reliable due to "noise," but are much more practical, making multispectral data preferable for accurate weed identification
GSD MSP: 5cm/pixel.
GSD RGB: 3cm/pixel.

Green-on-Green
For green-on-green weed detection, recent research indicates that high-resolution RGB cameras are particularly effective. Integrating RGB imagery with advanced identification tools enables reliable detection of weeds growing within crops, pastures, and other vegetative environments..
For accurate detection, one must fly at a time when the weeds are visually distinct from the crop (e.g., by color, height, or diameter).
The "Magic Tool" provides identification for all weeds that are visually different from the crop, but it does not classify them by type.
For weed identification, satellite imagery from Sentinel-2 is discarded due to its low resolution (10 meters per pixel).
Examples:
-
Sorghum halepense in peanuts:
- Sorghum halepense in wheat:
-
Broadleaf weeds in pastures
-
Volunteer crops