Abstract for presentation at The 13th Australasian Remote Sensing and Photogrammetry Conference

Grapevine canopy characteristics as determined from remotely sensed imagery have been shown to be effective in forecasting grape composition parameters that can be used to estimate the quality of wine made from those grapes. Maps of canopy characteristics are therefore valuable tools for precision viticulture practice. In a case of extracting reflectance data at the scale of individual vines from vineyard imagery with a pixel resolution of ca. 0.5 m, simple use of sample point location data provided by a GPS (the GPS points) proved too inaccurate for the desired analysis. At the individual vine scale, the spatial error between the GPS point and the corresponding location in a georectified image was great enough to result in clearly incorrect pixels being identified as representative of the sample grapevine canopy. The sample GPS point locations were, however, sufficiently close to the correct vine canopy to act as a seed point for a computer search algorithm.
The VineClipper algorithm was developed to identify a more representative set of vine pixels using the sample GPS points as seed points in a proximal spatial search. The procedure automatically recognises vine rows in an image close to each GPS point and then determines the local central line of the vine row closest to the GPS point location. The central point of the canopy pixels selected to correspond to the GPS field data point was selected as the point on the centre line closest to the GPS point. Reflectance data of the pixels surrounding this point were then extracted from inside a rectangle (equivalent to the area of one vine) aligned with the row direction. The methodology and the novel algorithm that was developed to automatically perform this procedure are presented.