Determination of weeds and artichoke plants position in colour images for local herbicide action

Abstract

Increased concern about environment is the main reason of developing European projects like AIR 1299 "PATCHWORK", aimed at reducing herbicide usage in cereal crops and eliminating chemical weeding in horticultural crops. The approach in horticultural crops is to design and build a robotic arm that will be able to locate weeds by machine vision and eliminate them by non-chemical means. A colour image analysis system has been developed and tested at the IVIA. The procedure starts with a segmentation of the RGB space in three categories (artichoke plants, weeds and soil) by means of Bayesian Discriminate Analysis. Because of some overlapping between classes crop and weeds, further processing is needed: filtering and object reconstruction. As a result of the procedure, a safety zone (region close to the plant boundary, in which weed control action is forbidden in order not to damage the crop) is defined and weeds are located in the image plane. First test showed encouraging results in images of top views of the plants that were slightly shadowed with an opaque surface.