Weed recognition

The automated identification of weed species is performed by machine-learning methods that allow species to be distinguished by their shape parameters.

Weed control plays an important role in agricultural production. As manual weed scouting is practically impossible for large-scale agriculture, the only solution now is to use herbicides, which significantly reduces quality of product, as well as increasing production costs. BitRefine group offers a solution for automatic weed detection and recognition based on modern computer-vision methods. This solution provides an opportunity to manage weeds in a way that is highly precise by removing them either mechanically or chemically.

Advantages of computer-vision weed control systems:

  • Reduce usage of chemicals
  • Improve crop quality
  • Improve crop quantity
  • Offer possibility of mechanical elimination of chemical-resistant weeds
  • Improve soil quality
  • Reduce overall costs

The base imaging installation consists of camera, artificial light source, and processing unit. The detection system installed on a vehicle and working at a speed of 50 images/second can control a weed-removal module in real time. As the accuracy of computer-vision classification heavily depends on training a set of image libraries, the close cooperation between BitRefine group as a computer science company and the farm is essential. Currently, the accuracy of weed recognition already exceeds 90%. However, the number of different weed species, variations in leaf shapes, and environmental conditions leave much room for further improvements.

Related Solutions

For more details, please contact us: