Discrimination of corn, grasses and dicot weeds by their UV-induced fluorescence spectral signature

Abstract

Real-time spot spraying of weed patches requires the development of sensors for the automatic detection of weeds within a crop. In this context, the potential of UV-induced fluorescence of green plants for corn-weed discrimination was evaluated. A total of 1 440 spectral signatures of fluorescence were recorded in a greenhouse from three plant groups (four corn hybrids, four dicotyledonous weed species and four monocotyledonous weed species) grown in a growth chamber. With multi-variate analysis, the full information contained in each spectrum was first reduced to the scores calculated from five principal components. Then, a linear discriminant analysis was applied on these scores to classify spectra on a species/hybrids basis and, subsequently, the resulting classes were aggregated according to the three plant groups. This two-step process minimized the error generated by heterogeneous groups such as dicotyledonous weeds. The output of this classification shows the significant potential of UV-induced fluorescence for plant group discrimination as the success rate reached 91.8%. No error was observed between corn and dicot weeds and most of the errors between corn and grasses came from confusion between the hybrid Pioneer 39Y85 and Setaria glauca L. (Beauv.). Analysis also determined that the position of the fluorescence sensor on the leaf and the plant age had negligible effects on the efficiency of fluorescence to discriminate plant groups. The factors to consider for transferring the results about UV-induced fluoro-sensing from laboratory to the field are discussed.