Detecting Weed Through Edge Computer Vision to Reduce Environmental Footprint
Challenge completed! Results follow soon.
Impact-driven startup Weedbot is developing a laser weeding machinery for farmers that can localize plants, distinguish between crops and weeds and remove weeds with a laser beam. 50 technology changemakers are building machine learning models to facilitate pesticide-free food production.
Laser weeding can substantially reduce or even eliminate the need for chemical herbicides, thus reducing soil and groundwater pollution. This would also facilitate pesticide-free food production and reduce the final price for such food, encouraging people to buy organic food and follow a healthy lifestyle.
The project goals
The goal is to develop a high-speed plant image recognition neural network with a speed of 12ms per image or faster and recognition precision of 100-110% of crop polygon (which means up to 10% false positives are allowed).
Sample image with annotated carrots:
Requirements for the AI
Input: high framerate video stream from the top of the crop row.
Necessary output: real-time high-speed weed segmentation with precision 1-2mm
Additional target requirements
Time must be not more than 12ms (from capturing the image till putting the converted lines to the User Datagram Protocol (UDP) queue)
The captured image must cover a 200x200mm working area with carrot seedlings as an object to be recognized.
The recognition software should recognize all carrots and up to 10% false positive (e.g. weed identified as “crop”) are acceptable.
The weed segmentation can be done either with the same AI that detects carrots or by a separate script like PlantCV.