Wheat plots segmentation for experimental agricultural field from visible and multispectral UAV imaging

Abstract

The use of Unmanned Aerial Vehicles (UAV) in precision agriculture (PA) has increased recently. Most applications capture images from cameras installed in the UAVs and later create mosaics for human inspection. In order to further improve the quality of data offered by this technology, application specific image processing algorithms that enhance, segment and extract information from the raw images delivering information equivalent to in-situ measurements are still missing. The present study describes a method for image segmentation to assist the characterization of nitrogen content in wheat fields. The proposed methodology uses the UAV and Computer Vision algorithms that process visual (RGB) and multispectral agricultural images. Data is first collected by the UAV that flies over an area of interest and collects high resolution RGB and multispectral images at a low altitude. Subsequently, a mosaic is created for each crop stage and the proposed algorithm segments the ROIs (regions where wheat crop is present) based on vegetation index. Using the proposed algorithm, the wheat plots are correctly segmented for two kinds of Brazilians wheat cultivates. The segmentation was validated by experts indicating that the proposed algorithm is suitable to be used as a first step of a method that assist the analyses of nitrogen content specific to wheat crops.