Crop-weed Discrimination by Line Imaging Spectroscopy

Two line imaging spectrometers covering the visible (VIS) and near-infrared (NIR) wavebands were further developed from remote areal application to close range application on single plants or parts of plants. Using artificial light, the line imaging NIR spectrometer was applied to record reflectance spectra in the wavelength range 660-1060 nm from small areas (1·5 by 6 mm) on young plants and background (soil) for the investigation of plant species and crop-weed discrimination. Spectra from sub-areas containing only pure plant surfaces of five different species were separated from others (soil, shadow or mixed) by the use of linear discriminant analyses in an extended segmentation process. Subsequently, linear and quadratic discriminant analysis, principal component analysis with soft independent modelling of class analogy, and partial least-squares regression with several Y -variables were applied in discrimination of crop and weeds on the reflectance characteristics. Of these methods, the bilinear methods, which are new within this research area, showed the highest classification performances of 70-80% on populations of four species (one crop and three weed species), and up to 90% when divided into two target groups (crop and weeds).