Detection of cherry tree branches with full foliage in planar architecture for automated sweet-cherry harvesting

Fresh market sweet cherry harvesting is a labour-intensive operation that accounts for more than 50% of annual production costs. To minimise labour requirements for sweet cherry harvesting, mechanized harvesting technologies are being developed. These technologies utilise manually-placed limb actuators that apply vibrational energy to affect fruit release. Machine vision-based automated harvesting system have potential to further reduce harvest labour through improving efficiency by eliminating manual handling, positioning and operation of the harvester and/or harvesting mechanism. A machine-vision system was developed to segment and detect cherry tree branches with full foliage, when only intermittent segments of branches were visible. Firstly, an image segmentation method was developed to identify visible segments of the branches. Bayesian classifier was used to classify image pixels into four classes - branch, cherry, leaf and background. The algorithm achieved 89.6% accuracy in identifying branch pixels. The length and orientation of branch segments were then analysed to link individual sections of the same branch together and to represent the branches with an equation. Linear and logarithmic model equations were fitted to the branch segments and the equation with minimum residual was selected as the best-fit model representing the corresponding branch. Branches detected with this algorithm were compared with manual counting. The method achieved a branch detection accuracy of 89.2% in a set of 141 test images acquired during full-foliage canopy. This study shows the potential of using a machine vision system for automating shake-and-catch cherry harvesting systems.