Assessing the potential of hyperspectral remote sensing for the discrimination of grassweeds in winter cereal crops

This article explores the potential use of remote sensing to discriminate two grassweeds (Avena sterilis and Lolium rigidum) from four cultivars (cvs) of winter wheat and barley. Hyperspectral measurements, using a GER2600 spectroradiometer (350 to 2500 nm), were conducted throughout the life cycle of the plants in order to analyse spectral differences between weeds and crops at different phenological stages. Specific techniques for hyperspectral data, such as the Spectral Angle Mapper (SAM) were used to quantify the spectral separability between weeds and crops, while stepwise discriminant analysis was applied to detect those wavelengths providing the best discrimination ability. SAM results showed that spectral differences were generally insufficient to discriminate weeds and crops. Only during the first phenological stages were angular distances large enough to achieve a good classification of the different species. This behaviour was related to the different fraction cover of crops and weeds in this period. The wavebands that provide the best discrimination ability according to the discriminant analysis were pooled in eight spectral regions in order to determine their frequency of occurrence. The four most frequently selected spectral regions were the Far Short-Wave Infrared, Early Short-Wave Infrared, Blue and the Red Edge.