63
IS, Vol. XXXVIII, Part 7B
In: Wagner W„ Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
[AN
)ATA
ite, Brazil,
e, Brazil
Hands
r, soil and vegetation
ports a miscellany of
[rounded by semiarid
quaqntities of organic
spectral and textural
v altitude photographs
s. Classification trials
oach. As a result five
iccuracy of over 80%.
s associée à l'eau, aux
ne et les populations
du Pandeiros est un
tés de macrophytes et
nous proposons une
le extraite à partir des
ihotographies à faible
graphiques. Des essais
évaluer l'approche la
atre classes terrestres
dioration significative
ts all this characteristics
site in Brazil. It is even
egion of water scarcity
It provides several
nutrients, fixing carbon
its relevance the State
> a Wildlife Sanctuary.
:d area does not ensure
areas still been used to
e drained and used by
¡a, 2009).
tying types of wetlands
subsidized its proper
theless, this task is a
plex ecosystem with a
The inventory of wetlands demand field surveys, aerial photo
interpretation and satellite imagery. Melack (2004) points out
that the use of satellite images is considered the most efficient,
since it allows a fast data acquisition and cartographic mapping.
A large range of tools are available to classify this sort of data.
However, high resolution images require more sophisticated
approaches. Texture has achieve expressive results in the
classification of those images. For example, Davis et al (2002)
obtained an overall accuracy classification of 75% using image
texture for riparian zones. Thus, as a starting point, we decide
evaluate texture potential of classify different groups of aquatic
plants based on Ikonos images.
1.1 Gray Level Co-occurrence Matrix
Several methods can be applied to texture analysis. Among
these, the Gray Level Co-occurrence Matrix (GLCM) seems to
be the most commonly used (Franklin, 2001) and has been
recognized as one of the best tools for specific situations of
classification (Clausi, 2000; Maillard, 2003). GLCM is a second
order histogram in which each entry reports the join probability
of finding a set of two grey levels at a certain distance and
direction from each other over some pre-defined window
(Maillard, 1999). Haralick et al (1973) was the first to extract
texture features in order to classify images. 14 textures
measures were originally described by Haralick. However many
features are highly correlated which made five of them more
popular: Contrast, Angular Second Moment, Entropy, Inverse
Difference Moment and Correlation.
In this study, we aimed to evaluate the use of GLCM in the
classification and segmentation of high resolution image of a
wetland environment. As well as determining the optimal
parameters of textures, window size and distance to be used in
the study of IKONOS images for this sort of environment.
2. METHODOLOGY
2.1 Study Area
Figure 1. Location of Pandeiros Wildlife Sanctuary in Minas
Gerais (MG) - Brazil.
The Pandeiros Wildlife Sanctuary (PWS) is located near the
Pandeiros’ River mouth, in the Northern part of the State of
Minas Gerais (Figure 1). This river is an important affluent of
Sao Francisco River and is the breading grounds of several
species of fish. It is also a refuge for numerous rare endemic
and threatened bird species (Biodiversitas, 2005). The region is
protected by the State government authorities and is managed
by the Forest Institute of Minas Gerais (IEF-MG).
It occupies a total area of 6103 ha. and preserves a unique
wetland with riparian forests, palm swamps, wet meadows,
lakes and ponds (Figure 2). Climate presents two distinct
periods: wet season from October to March, and dry from April
to September. This variation is characteristic of the Cerrado
biome where water deficit spans for about half the year.
2.2 Field Work
The first of Four field campaigns was conducted in September
2008 using a boat and an all-terrain vehicle to access difficult
areas for a general reconnaissance approach. During the second
one in February 2009 geodetic ground control points were
collected for the geometric correction and registration of the
image. A specific work area was also defined and data was
acquired on the different vegetation physiognomies that could
be identified on the Ikonos image. The third campaign in May
2009 was mainly dedicated to acquire low altitude photographs
using a micro-light aircraft to serve as complementary
validation data. The Fourth and last one in April 2010 allowed
acquiring new low altitude photographs and visit a few spots
where some botanical inventory was still necessary. During the
last three campaigns, printed copies of the Ikonos image (scale
1:5000) were used to identify complexes both in the field and
on the image. This data allowed us to divide vegetation of the
study area in 9 different classes (Figure 2): Pontederiaceae,
Nymphaeaceae, Riparian forest, Open Water, Alismataceae,
Cyperaceae, G - Pasture, H - Flooded Pasture, I - Bare Soil.
Figure 2. (a) Fusionned, false color Ikonos image of the
Pandeiros. The image represents an area of 1200x1200 pixels or
144 ha. Legend: A - Pontederiaceae', B - Nymphaeaceae', C -
Riparian forest', D - Open Water, E - Alismataceae F -
Cyperaceae', G - Pasture', H — Flooded Pasture', I - Bare Soil.
Photographic records of different vegetation typology were
acquired to constitute a visual inventory of the Pandeiros. The
dominant plant groups present on the photographs were
identified by two botanists at the Botanic Taxonomy
Laboratory of the Universidade Federal de Minas Gerais
(UFMG). The aerial photographs proved to be useful for the
inventory and as validation data. Since only a navigation GPS
was used in the last two campaign, care was taken to note and