On Feature Extraction for Fingerprinting Grapevine Leaves

摘要

Within the scope of CROP.SENSe.net, an interdisciplinary research network of Bonn University and the Julich Research Centre, we work on a new model-based approach to the phenotyping of grapevine. Our algorithm performs a robust extraction of different features from a given leaf image, like specific points of the vein network, the vein network itself, and different distances respectively angles between special features. For that we present robust methods, like a template based method to extract the peduncle point, a detection strategy to determine end points of leaf veins, and a Gabor filter-based directional edge tracing procedure to extract the network. The extracted features are fed into a support vector machine in order to realize a full automatic sufficient variety identification.