Project Summary
WeedElec: combining artificial intelligence, drones and robotics for herbicide-free weeding
The WeedElec project explores a new approach to precision weeding based on the combination of artificial intelligence, aerial imaging, robotics and electrical weed control. Supported by the French National Research Agency (ANR) through the 2017 ROSE Challenge, the project brings together expertise in agricultural robotics, imaging, artificial intelligence, botany and plant physiology.
Its goal is to develop the foundations of a system capable of detecting weeds, automatically identifying them and targeting only the plants that need to be removed. The system combines a drone, an agricultural robot equipped with a fast articulated arm and a high-voltage electrical weeding tool.
Pl@ntNet for identifying weeds
Within this system, Pl@ntNet contributes to the development of automated plant identification methods for agricultural environments. Unlike conventional photographs taken by users, images captured by drones and robots may show very small weeds, partially hidden plants or plants viewed from unusual angles.
Pl@ntNet teams therefore explored how plant recognition and deep-learning methods could be adapted to these new observation conditions. Several approaches were investigated to adapt recognition models to automatically acquired images and enable them to run on robots with limited computing resources.
This research broadens the potential uses of Pl@ntNet: recognition technologies originally developed for participatory plant identification can also be adapted to automated sensors and robotic systems capable of acting directly in the field.
From detection to targeted intervention
Drone imagery can be used to map infested areas and prepare targeted interventions. Once on site, the robot uses its onboard sensors to refine weed detection and accurately position the weeding tool.
The system then applies a high-voltage electrical pulse directly to the targeted plant. WeedElec investigates the electrical parameters required to achieve effective weed control while limiting energy consumption and protecting crops.
The approach takes into account several factors, including plant size and growth stage, water status, as well as soil characteristics.
Research at the intersection of botany and precision agriculture
By combining aerial observation, plant recognition, robotics and targeted intervention, WeedElec explores an approach based on a simple principle: observe before acting, identify the plant and intervene only where necessary.
For Pl@ntNet, the project provides a research environment at the intersection of botany, artificial intelligence and precision agriculture. It contributes to exploring new conditions for automated plant recognition and developing technologies that could help reduce reliance on herbicides.
Programme: ROSE Challenge, 2017 edition
Funding: French National Research Agency (ANR)
Fields: agroecology, artificial intelligence, plant recognition, imaging, drones, agricultural robotics, electrical weeding.
WeedElec News
WeedElec: where to find news on the project?
The WeedElec project has an official website, which can be accessed HERE.
