ETUDE COMPARATIVE DE METHODES DE CLASSIFICATION
POUR LES SOLS ET LES VEGETAUX
C. Morzier, P. Meylan, A. Musy
Institut de Génie Rural
Ecole Polytechnique Fédérale de Lausanne (Suisse)
Abstract
Presentation of the complete data processing system of numerical
pictures for soils and crops classification. Three different data sets are
used for these studies: ground truth data for soil and vegetation, remote
sensing data collected by an areoplane multispectral scanner, and areal
photographies. The multispectral data are controlled by different statis-
tical methods and with a visual examination of a computer print-out. The
conventional photographies are digitized with an electro-optical camera
connected to a mini-computer.
The software package of the AEI includes several classification
programs using supervised and unsupervised technics. For a supervised
classification, the sample processing is done with ground truth data,
visualisation and areal photographies. The quality of the sample technic
is increased by using a descriptive discrimination based on a hierarchical
clustering and a decisional discrimination of the samples. Three supervi-
sed classifications schemas are used: Sebestyen's classification, rectan-
gular and barycentric methods. The results of these classifications are
compared to the ground truth data of several verifications test field. So,
it is possible to compare the performances of the algorithms used.