Discrimination of pure grassland species using NIR Hyperspectral Imaging

The objective of this study was to discriminate by hyperspectral imaging system, SWIR ImSpector N25E, different pure grassland species (Festuca rubra L., Trifolium repens L., Agrostis capillaris L., Hieracium aurantiacum L., Arnica montana L.) into grassland species mixtures. All the samples were collected from natural meadows of the National Apuseni Park, Apuseni Mountains, Gârda area (Romania). The samples were air-dried, then prepared using the protocol for NIRS analysis adapted on the scanning linear system. For images acquisition, the Hyper See program was used. Then a model build under MatLab (PLS–DA) was used to discriminate pure species from the mixtures of two or three species. This analysis was carried out in order to see, on images obtained previously from the floristic composition of experimental parcels, if the pure species are or are not recognized according to the spectral data base. More than 99% correct predictions for species discrimination were obtained. This study should guide us to verify if a toxic species is present or not in natural meadows used as food for animals. The floristic composition of a meadow can be determinate only if we have in the data base, spectra for each identified species, as being part of the mixture.