u 2004
cessing
'rentice
Boston:
on and
lishing
logical
. Image
tomatic
a Shift-
esearch,
it 96 A,
ygnition
ce and
Tehran
omatics
Remote
mmetry,
emotely
nposium
wa.
Itisensor
ire and
vidence.
n — the
of the
mation
on data
FUSION OF MULTISENSOR REMOTE SENSING DATA:
ASSESSING THE QUALITY OF RESULTING IMAGES
E. Saroglu, F. Bektas, N. Musaoglu, C. Goksel
a
ITU, Civil Engineering Faculty, 34469 Maslak Istanbul, Turkey Istanbul
saroglue@itu.edu.tr, bektasfi@itu.edu.tr
Commission IV, WG IV/7
KEY WORDS : IRS, Landsat TM, SPOT, Radar, Fusion techniques, Land cover, Land use, Accuracy.
ABSTRACT:
The primary attention of this study was to examine what improvement can be obtained for classification accuracies by using
different merging techniques done with multisensor dataset. In this study, the existing fusion techniques that preserve spectral
characteristics, while increase spatial characteristic such as Principle Component Analysis, Intensity-Hue-Saturation, Brovey and
Multiplicative algorithms were applied to multi sensor data set. IRS 1 D Pan, LISS III and ERS images were used. Using fusion
techniques IRS 1 D imagery combined with LISS III data and ERS radar data combined with LISS III remotely sensed data.
Maximum Likelihood classification algorithm was applied to classify fused imageries. Before classification procedure training sites
were selected for all various land cover/use categories. Classification accuracy assessment was calculated using an error matrix for
all images. Finally, the results of classification accuracy were compared and the best result was obtained by combining IRS 1 D
image with LISS III data by means of IHS colour transformation technique.
I. INTRODUCTION
Wald (2002) describes fusion as ‘a formal frame work in which
are expressed means and tools for the alliance of data orginating
from different sources. It aims at obtaining information of
greater quality; the exact definition of greater quality will
depend upon application’. The data fusion of multisensor data
has received tremendous attention in the remote sensing
literature (Yao and Gilbert, 1984; Welch and Ehlers, 1988;
Chavez et al., 1991; Weydahl et al., 1995; Niemann et al., 1998;
Pohl et al, 1998; Saraf, 1999; Zhang, 1999; Gamba and
Houshmand, 1999). The integration of spectrally and spatially
complementary remote multi sensor data can facilitate visual
and automatic image interpretation (Zhou et al, 1998). Data
fusion is the combination of multi source data which have
different characteristics such as, temporal, spatial, spectral and
radiometric to acquire high quality image. The fusion of
different sensor images is crucial method for many remote
sensing applications such as land cover/ land use mapping.
There is a huge variety of techniques to combine images from
different sensors. However, this paper focuses on image fusion
techniques that preserve spectral characteristics whilst
increasing spatial resolution to provide images of greater
quality. IRS 1 D Pan, LISS III and ERS images were fused by
using Brovey, Multiplicative, IHS and PCA algorithms. All
merged images were classified by means of Maximum
Likelihood supervised classification technique. The overall
accuracy and Kappa analysis were used to perform a
classification accuracy assessment based on error matrix
analysis. The quality of the fused images was examined by
comparing classification accuracy results.
2. STUDY AREA
In this study, the region which is located southwest of Istanbul
between 41 6' 13"- 40 55' 36" latitude and 28 37" 3"- 28 54' 11"
longitude was selected as study area (figure 1). It comprises
approximately 400 km? area which contains the diversity of
land cover types and surface materials such as urban-built up,
vegetation, water, agricultural field, Transit European
Motorway and Yesilkôy Atatürk Airport. Land use in study site
is very cosmopolitan and irregular. Most buildings in the area
are small, form closely and do not have roof. The streets in the
city are narrow.
beet
MIDDLE
EAST
Figure 1. Study area