Crop weeds detection (Green on Green) - PIX4Dfields

PIX4Dfields supports green-on-green detection through high-resolution imagery and the Magic Tool. Follow this article to detect weeds within crop fields efficiently.

Introduction

Green-on-green detection in agriculture allows for identifying weeds among crops, which allows spot-spraying and minimizes blanket herbicide applications. By targeting only the affected areas, this method not only reduces herbicide costs but also minimizes environmental impact.

GSD Requirements

Orthomosaics with a GSD between 0.5 and 3 cm are ideal for detailed, plant-level weed detection. This ensures high-resolution detection of individual weeds within crop canopies.
If weeds occur in larger patches or zones, a lower resolution of GSD between 4 and 8 cm is generally acceptable.
The following image compares a flight at 20 mts with 0.5cm GSD on the left with a flight at 40 mts with 2.1 cm GSD on the right.

Consider weed condition

Weed growth stage affects detection efficacy:
Early stage: Green weeds blend with crops in the early stages and require higher image clarity for accurate identification.
Advanced stage or treated weeds: Weeds that are more mature or have been treated with herbicides may appear yellowish, improving detection due to color contrast.

Green-on-green detection example use cases

PIX4Dfields has been successfully used in various scenarios for green-on-green weed detection. Here are a few example use cases that highlight the method's effectiveness in detecting weeds amidst crops:

Pasture infested with Bitter dock (Rumex obtusifolius)

Orthomosaic size	2.66 GB
Average Ground Sampling Distance (GSD)	0.5 cm / 0.19 in.
Area covered	1.75 ha / 4.37 acres.
Image acquisition plan	1 flight, grid flight plan.
Drone and camera	DJI Mavic 3 Multispectral, RGB camera
Reflectance panel/target	Not applicable
Download example Orthomosaic

Download and import Orthomosaic to PIX4Dfields

Download the example layer from BitterDock_20mts_RGB.

To import an Orthomosaic into PIX4Dfields:

1. On the dashboard, click + New Project or the “plus” symbol at the bottom right corner of the window.
   
2. A new window with a background satellite image appears, displaying either the default location or where the last project was processed.
3. Click the pen symbol beside the default name of the project to rename the project.

To import the .tiff orthomosaic:

1. Under Other map, click GeoTIFF.
   
2. Select the example file "BitterDock_20mts_RGB.data" and click Open.
3. The Orthomosaic is imported successfully.

Detecting weeds with the Magic Tool

• Before running the weed detection, a boundary should be created.
  1. Click the Boundaries layer on the left board above the layer list to enter the boundaries mode.
  2. To draw a field boundary, click the Draw boundary icon in the left corner of the map.
  3. Select Boundary. Note that obstacles can be created as well.
  4. Left-click to start drawing.
  5. For more information: Field boundaries and obstacles - PIX4Dfields.

• Access the tool by clicking the Magic icon next to the Index icon.

• Select the Layer and Boundary.
• Set the Cell Size
  Your GSD influences the minimum cell size. A very small cell can degrade selection performance.
  Recommendation: Adjust the cell size to your crop and weed size, with values often ranging from 0,5 to 1 meter for high-resolution detection. 
  In this case, set it to 1 meter. 
• Labeling
  Manually identify areas of weeds and crops by clicking on them within the boundary:
  • Weeds: Click on at least 10 cells representing weed areas.
  • Crops: Click on at least 10 cells representing crop areas.
  • Purpose: These sample points help the Magic Tool distinguish weeds from crops based on visual patterns and color differences
  • 
• Clean up and convert to Operation
  In Step 4, clean up the selection manually and add or delete cells.
  Later convert the results into an operation layer. This finalizes the weed detection, creating an operational layer that can guide targeted herbicide applications or further analysis.

Other Green-on-Green use cases! 

Peanut infested with Sorghum halepense.

A DJI M3E drone flew at a 2 cm GSD. Thanks to the high-resolution imagery and contrast between plant types, the Magic Tool helped identify the Sorghum halepense plants within the peanut crops relatively straightforwardly.

The Magic tool cell size was around 2 meters.

Rice Infested with Sedges (Cyperaceae family)

Using a DJI Mini 3 Pro, a flight was conducted at 1.5 cm GSD over rice fields infested with Sedges. The Magic Tool enabled the precise detection of sedge patches within the rice field. This detection accuracy facilitates efficient, targeted treatment of sedge-infested areas.