Resumen de Hyperspectral imagery to discriminate weeds in wheat
Gilles Rabatel, Farida Ougache, Nathalie Gorretta, Martin Ecarnot
The problem of weed and crop discrimination by computer visionremains today a major obstacle to the promotion of localized weeding practices.The objective of present study was to evaluate the potential of hyperspectralimagery for the detection of dicotyledonous weeds in durum wheat during weedingperiod (end of winter). An acquisition device based on a push-broom cameramounted on a motorized rail has been used to acquire top-view images of crop at adistance of one meter. A reference surface set in each image, as well as specificspectral preprocessing, allow overcoming variable outdoor lighting conditions. Thespectral discrimination between weeds and crop, obtained by PLS-DA, appearsparticularly efficient, with a maximal error rate on pixel classification lower than2%. However complementary studies addressing robustness are still required
