What is the challenge that is being addressed?

Detection and treatment of weeds constitutes a major challenge for the agricultural industry with the expenditure on crop protection chemicals accounting for up to 10% of the total cost of production and potential yield-reduction up to 80% from improper detection and treatment. Additionally, crop protection chemicals, such as Glyphosate present a major threat to the environment and biodiversity. Farmers spend extensive efforts and resources on manual inspection of the fields in order to allow for early detection and treatment of weed infestations. Correspondingly, large scale automated detection of weeds would form a major contribution to both economic efficiency and environmental sustainability of agricultural production.

What is the AI solution the project has implemented?

The AI solution which has been developed is an automatic, large-scale detection of in-field weed detection using super-resolved Sentinel-2 at 1m per pixel resolution and automatically delineated field boundaries (in-season). Additionally, the AI model focuses on weeds in pre-closed-canopy stage and with a minimum patch size of 3x3 meters. The model developed reached an accuracy of 92% across over 500k hectares in Brazil and is able to process 1.sq.km of data per second.

Who helped implement the AI solution?

This solution is implemented in the context of SR4C3, a winning project from the AI4Copernicus 1st Open Call. The consortium consists of DigiFarm, Altyn, and Farmen Gard, while the results of the project are available here.