Ground-level hyperspectral imagery for detecting weeds in wheat fields

Site-specific weed management can allow more efficient weed control from both an environmental and an economic perspective. Spectral differences between plant species may lead to the ability to separate wheat from weeds. The study used ground-level image spectroscopy data, with high spectral and spatial resolutions, for detecting annual grasses and broadleaf weeds in wheat fields. The image pixels were used to cross-validate partial least squares discriminant analysis classification models. The best model was chosen by comparing the cross-validation confusion matrices in terms of their variances and Cohen's Kappa values. This best model used four classes: broadleaf, grass weeds, soil and wheat and resulted in Kappa of 0.79 and total accuracy of 85 %. Each of the classes contains both sunlit and shaded data. The variable importance in projection method was applied in order to locate the most important spectral regions for each of the classes. It was found that the red-edge is the most important region for the vegetation classes. Ground truth pixels were randomly selected and their confusion matrix resulted in a Kappa of 0.63 and total accuracy of 72 %. The results obtained were reasonable although the model used wheat and weeds from different growth stages, acquisition dates and fields. It was concluded that high spectral and spatial resolutions can provide separation between wheat and weeds based on their spectral data. The results show feasibility for up-scaling the spectral methods to air or spaceborne sensors as well as developing ground-level application.