Nowadays, remote sensing is an essential science in French Polynesia because of its extended territory and the
remoteness of its 120 islands. There is a strong need to study the vegetation cover and its evolution (biodiversity
threat, invasive species, etc.).
A growing satellite images database has been acquired throughout, giving access to very high resolution optical
images such as Quickbird data. These data allow accessing the vegetation canopy spectral and contextual information,
texture classification has proved to be an efficient tool to map the complex vegetation found in tropical
The main goal of this paper is to propose an optimized SVM multispectral-texture classification method for
tropical vegetation mapping.
One of the texture computation drawbacks is the window treatment size, which is related to the largest texture
element size. In complex tropical vegetation cover, this parameter leads to very small ground truth learning
database, inducing a significant degradation of the classifications accuracy. We propose to increase the thumbnail
numbers using an under-sampling method, optimizing the size and the number of the thumbnails.
The other drawback is the high dimensionality of the problem when dealing with multispectral textures. We thus
propose to rank and select the most pertinent textures attributes in order to reduce the dimensionality without
reducing the classification accuracy.
We first introduce the study context, before exposing preliminary studies on tuning the SVM learning method.
The adapted method is then accurately exposed and the interesting experimental results as well as a sample of
applications are presented before to conclude.
The purpose of this study is to classify the types of coconut plantation. To this end, we compare several classifiers
such as Maximum Likelihood, Minimum Distance, Parallelepiped, Mahalanobis and Support Vector Machines
(SVM). The contribution of textural informations and spectral informations increases the separability of different
classes and then increases the performance of classification algorithms. Before comparing these algorithms, the
optimal windows size, on which the textural information are computed, as well as the SVM parameters are first
estimated. Following this study, we conclude that SVM gives very satisfactory results for coconut field type
The goal of this study is to classify the coconut fields, observed on remote sensing images, according to their
spatial distribution. For that purpose, we use a technique of point pattern analysis to characterize spatially a
set of points. These points are obtained after a coconut trees segmentation process on Ikonos images. Coconuts'
fields not following a Poisson Point Process are identified as maintained, otherwise other fields are characterized
as wild. A spatial analysis is then used to establish locally the Poisson intensity and therefore to characterize
the degree of wildness.
The Support Vector Machine (SVM) algorithm is assessed for the classification of polarimetric radar data for the
cartography of natural vegetation. Fully polarimetric data has been acquired in L and P bands during an AIRSAR
mission over the French Polynesian Island named Tubuai. The results show significant improvement when compared to
those obtained with the classification based on the maximum likehood criterion applied to the theoretical Wishart
distribution that are supposed <i>a priori</i> to be verified by radar data. Obviously, this hypothesis is not verified with the
present experimental data over the study site. The addition of other polarimetric indicators to the elements of the
polarimetric coherency matrix still improves the classification accuracy. The evaluation of different partial polarimetric
modes shows that even the best results are obtained for fully polarimetric data, the π4 mode gives the best compromise
with respect to the ASAR Alternate Polarization mode or the PALSAR Dual Polarization mode. This latter shows in turn
better results than the Alternate Polarization mode, indicating the significant contribution of the polarimetric differential
phase between 2 polarization channels.