Feasibility of a real-time weed detection system using spectral reflectance
In: Precision Agriculture ‘05Search for other papers by J. Bossu in
Current site
Google Scholar
PubMed
c.gee@enesad.fr
Search for other papers by Ch. Gée in
Current site
Google Scholar
PubMed
Search for other papers by J.P. Guillemin in
Current site
Google Scholar
PubMed
Search for other papers by F. Truchetet in
Current site
Google Scholar
PubMed
Purchase instant access (PDF download and unlimited online access):
For site specific herbicide spraying, the feasibility of a real-time weed detection system using spectral reflectance was studied. Using a spectrometer (spectral range : 400-1100nm, optical resolution : 7nm) as a diagnostic tool, leaf reflectance experiments were carried out either in the field or in the lab, under a controlled halogen-tungsten light source. Four local weeds have been studied : 1) creeping thistle (Cirsium arvense (L.) Scopoli), 2) common field speedwell (Veronica persica L.), 3) charlock (Siniapis arvensis L.), 4) wild-oat (Avena fatua L.), the only monocotyledon.
Using artificial neural networks (ANN) as supervised classification methods, we discriminated weeds from their reflectance. First, classification has been applied for discriminating vegetal groups : monocotyledons and dicotyledons. Then, a more precise classification, based on weed species, has been investigated based on a multi-layer perceptron (MLP). The classification results are presented and the choice of these classification models is discussed. The field results indicated that weed reflectance is not reliable for managing weed detection in real time. We are presently considering the use of geometric information provided by a CCD camera to improve the classification performance.