Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Detection of visual grape leaf characteristics for automated ampelography type

Tsellou Aikaterini

Full record

---

Summary

Ampelography is the branch of viticulture that studies the description, distinction, classification and evaluation of grapevine varieties. In the modern era of varietal wines, correct identification of different grapevine varieties is necessary as it can have a substantial financial impact on the wine industry. The development in digital photography and image processing tools offers enhanced capabilities for ampelography by providing automated and more accurate methods to discriminate leaves, replacing the classic technique. In this thesis, we prove that machine learning algorithms are able to classify efficiently different kinds of grape leaves in an automated way. The proposed approach consists of the following phases: segmentation, feature extraction, feature selection and classification. In the segmentation phase the leaf is separated from its background. Then, in the feature extraction phase, the segmented leaf image is analyzed in order to extract shape and contour features. After extracting the features, we select an optimal subset of them in order to perform the classification in the next phase. Finally, the results are classified using 3 different algorithms: Naïve Bayes, Decision Tree, SVM with linear kernel and quadratic kernel. Evaluating the classification results, it should be noted that the automatic extraction of morphological data and machine learning modelling proved to be rapid and accurate methods for cultivar classification.

Available Files

Services