EXTRACTION ET CARACTERISATION DES VIGNES A PARTIR DE DONNEES DE TELEDETECTION A TRES HAUTE RESOLUTION SPATIALE. APPLICATION EN LANGUEDOC-ROUSSILLON POUR LA CONSTITUTION DE BASES DE DONNEES GEOGRAPHIQUES.

摘要

This work aims to meet the vineyard managers' needs for information, using very high spatial resolution aerial images. Three methods have been developed and compared for vine plot detection. They are based on oriented and periodical texture recognition using either a cooccurrence analysis (Haralick's contrast index) or a frequency analysis (Fourier transform and Gabor's filters). The Fourier transform calculation on a sliding window provides the best results with plot segmentation in polygons as well as a precise estimation of row orientation and interrow distance. These characteristics enable the extraction and detailed study of each vine row, designed to: 1) improve definition of plots contours, 2) detect missing vine plants and 3) characterize inter-rows. Most of the users' needs have been met during this PhD study thanks to the implementation of a semi-automatic tool for vine plot detection, segmentation and characterization.