Thresholding Procedures for Bare Soil / Herbaceous Vegetation Discrimination

To separate soil from an herbaceous crop, winter wheat, this work applies the multilevel OTSU algorithm to two vegetation indices derived from UAV multispectral and airborne hyperspectral images in three different phenological periods. NDVI and ChRE (Chlorophyll Red Edge) spectral indices have been used for this purpose. Linear regression has been carried out to predict crop traits such as yield, nitrogen content, and biomass. This work has shown that the application of multilevel thresholding in a spectral space represented by a vegetation index offers a suitable strategy to identify the optimal threshold to separate soil from crop. The best results have been achieved by applying a threshold obtained from a 4-level threshold procedure during the middle growing period with an NDVI calculated from airborne hyperspectral bands. In this case, 0.60, 0.69 and 0.68 R2 values were obtained, respectively, with yield, nitrogen content and biomass measurements.