Résumé
The project ‘WeedElec’ aims at developing a robotic intra-row weeding solution based on a high-voltage selective electrocution. It performs the following tasks: weed detection and location, positioning of the weeding probe according to location results and electrical destruction of weeds. The detection is achieved using an RGB camera associated with the deep-learning solution of the ‘Pl@ntNet’ team. Discrimination results between crops and weeds attained 76% accuracy for maize and 62% for beans (field trials). Mortality rates weeds ranged from 75% (Polygonaceae) to 13% (Brassicaceae) for plants less than one month old (laboratory conditions). Survivors were nevertheless impacted with a severe limitation of their growth.