IMAGE-ANALYSIS FOR PRUNING OF LONG WOOD GRAPE VINES

Image analysis algorithms were developed for making measurements relev ant to the winter pruning of long wood grape vines, and were tested on 60 images. The measurements required were, for each branch, the direc tion, diameter and length, and the position of the bottom end relative to the trunk axis and the lowest supporting wire. Each image was capt ured using a powerful flash which eliminated the effects of the ambien t daylight. The flash was tilted downwards by approximately 15 degrees to the horizontal because the unreflective trunk required more illumi nation than the branches. Segmentation of the images was by thresholdi ng, size-filtering for noise removal, and reduction to the skeleton. L oss of grey level information and distortion of the skeleton in the re gion of junctions and crossovers created problems when resolving the v ine structure. The wire was located by a Hough transform and the trunk axis was located using a histogram method. The wire was located succe ssfully in 95% of the images. The trunk axis was located successfully in all of the images. Of the primary branches visible in the images, 8 9% were found by the image analysis, and 80% of these were drawn corre ctly connected together. The positions of the bottom ends of the branc hes were not measured accurately enough for pruning purposes, but coul d have been extrapolated to their correct positions if the location of the long wood had been known with more precision. (C) 1997 Silsoe Res earch Institute.