Counting vine shoots early in the growing seasonis critical for adjusting management practicesbut is challenging to automate due to arange of environmental factors. This paper proposesa complete framework for shoot detection,comprised of image preprocessing, featureextraction and unsupervised learning as a -nal clustering step. Experiments on four vineblocks across two cultivars and training systemswere conducted. The results showed the overallframework was successful at detecting shootsand in particular was robust to a range of lightingconditions and other environmental impactsthat limit the success of prior work. This frameworklays the foundations for full automation ofshoot mapping on a large scale in vineyards.
展开▼
