Goal
The goal is to develop a general foundational deep-learning model that detects weeds within crops. The model utilizes high-resolution drone imagery to generate centimeter-level weed and crop maps. These maps will enable farmers to produce accurate spray maps for precision spot-spraying, significantly reducing herbicide usage, environmental impact and improving farmers’ competitiveness.
Plan
The plan for the project is to accumulate vast amounts of data of relevant crop and weed combinations in all Northern Europe, with a high amount of variation in all aspects of drone, flight, camera and crop/weed growth.
The images must then be accurately annotated into relevant plant categories to allow for meaningful insights.
A large general foundational model must then be developed with relevant state-of-the-art (and beyond) deep-learning practices to ensure high robustness.
Finally, the models must be implemented in a user-friendly setting, tested and released to the market.
Expected results
The expected result is an AI model implemented in an intuitive and user-friendly platform, where farmers, drone pilots and agronomists can simply upload drone images, and the model will automatically predict what crop it’s looking at and what/where weeds are in the field. Results can then be used to generate flexible spray maps that fit existing equipment and reduce herbicides by up to 95%.
