Automated detection and segmentation of vine rows using high resolution UAS imagery in a commercial vineyard

Climate models predict increased average temperatures and water scarcity in major agricultural regions of Australia over the coming decades. These changes will increase the pressure on vineyards to manage water and other resources more efficiently, without compromising their high quality grape production. Several studies have demonstrated that high-resolution visual/near-infrared (VNIR) vineyard maps acquired from unmanned aerial systems (UAS) can be used to monitor crop spatial variability and plant biophysical parameters in vineyards. However, manual segmentation of aerial images is time consuming and costly, therefore in order to efficiently assess vineyards from remote sensing data, automated tools are required to extract relevant information from vineyard maps. Generating vineyard maps requires separating vine pixels from non-vine pixels in order to accurately determine vine spectral and spatial information. Previously several image texture and frequency analysis methods have been applied to vineyard map generation, however these approaches require manual preliminary delineation of the vine fields. In this paper, an automated algorithm that uses skeletonisation techniques to reduce the complexity of agricultural scenes into a collection of skeletal descriptors is described. By applying a series of geometric and spatial constraints to each skeleton, the algorithm accurately identifies and segments each vine row. The algorithm presented here has been applied to a high resolution aerial orthomosaic and has proven its efficiency in unsupervised detection and delineation of vine rows in a commercial vineyard.