Decomposition of Mixed Pixels of ASTER
Satellite Data for Mapping Chengal
( ) Tree
Noordyana Hassan
1
, Mazlan Hashim
2
!" # $%
1
noordyana2@live.utm.my
2
mazlanhashim@utm.my
&’—
! "#$
%
&
&&’ (
$
)
*
+,- +..
/,,,0 "#$
I. INTRODUCTION
An understanding of Mixture Tuned Match Filtering
(MTMF) can minimized difficulty in recognizing patterns of
mixed pixels (mixels) that usually occurred in highly
heterogeneous scene such a tropical rainforest. Amongst one
the recent widely used technique to decompose these mixels is
MTMF, particularly reported for forestry applications [1].
MTMF is a spectral unmixing algorithm where measured
spectral of mixed pixel is decomposed onto a collection of
constituents spectra or endmembers and a set of constituents
spectra or abundance that indicates the proportions of each
endmember present in any given pixel [2].
There are three parameters that contribute to mixed
pixels problem: (i) low spatial resolution; (ii) feature
homogeneity; and (iii) low spectral resolution. High spectral
resolution is appropriate for mapping individual timber
species with high precision and accuracy. This is because
differentiation of individual timber species may cause some
problem because trees within same genera have somewhat
similar spectral characteristics. Thus, spectral detail is
necessary for distinguishing similar features [3]. High spatial
resolution also contributes to higher accuracy in automated
sub-pixel classification. Low spatial resolution may result
mixed pixels because tropical rainforest was heterogeneous,
highly complex, with higher density of emergent and tall
canopy trees [4, 5, 6]. Radiometric resolution plays important
roles in sub-pixels classification. High radiometric resolution
may increase ability to distinguish fine differences in
reflectance values among pixels [7], such a case is rare in the
high heterogeneity and complexity of the tropical rainforest
Various approaches to digital image classification of
satellite remote sensing data have been utilized for classifying
satellite images for forestry applications. This includes
pattern recognition algorithms [8], fuzzy [7] and neural
network based algorithms [9]. However, the results from these
classifications were found very low and far from expectations
at operational level due to occurrence of high proportion of
mixels within the scene. In mapping-related industries,
acceptable operational accuracy of >85% for overall average
classification is required or relationship (r
2
) of detected
features against reference data is > 0.85. Thus, this study is
carried out to apply MTMF on ASTER dataset by focusing on
heterogeneous area to minimize mixels before decomposed
them into their respective class proportion. The focus of this
study is to assess the utility of MTMF for estimating Chengal
trees composition in mixels over the scene. MTMF is the
fraction of an image which allows false positives to be
identified and eliminated from the abundance results. As such,
we can hypothesized that MTMF can be used to identify tree
species in the ASTER dataset as it can calculate the quantity
of target that much smaller than pixel size.
II. MATERIALS AND METHODOLOGY
&( ) *
The study area was located at Pasoh Reserve Forest (2°
58’N, 102° 18’E) in the state of Negeri Sembilan and 70 km
southeast Kuala Lumpur Malaysia. The study area that been
covered about 50 Ha of Pasoh Forest Reserved (Fig. 1). The
plot is a +", rectangle 1 - long by 0.5 - wide. The
vegetation is primary rain forest. The upper canopy is
dominated by red meranti, section Muticae, especially
2011 IEEE International Conference on Control System, Computing and Engineering
978-1-4577-1642-3/11/$26.00 ©2011 IEEE 74