International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Table 1. Sensor Characteristics - Landsat 7 ETM+ and SPOT 5 HRG
Wavelength Spectral Range (um) Spatial Resolution (m)
Region (with band no.)
Landsat 7 ETM+ SPOT 5 HRG ETM+ HRG
VIS Blue 0.45 - 0.515 (BI) - 30 :
VIS Green 0.525 - 0.605 (B2) 0.50 - 0.59 (Bl) 30 10
VIS Red 0.63-0.69 (B3) 0.61 - 0.68 (B2) 30 10
NIR 0.75-0.90 (B4) 0.78 - 0.89 (B3) 30 10
MIR/SWIR 1.535-1.75 . (BS) 1.58 - 1.75 (B4) 30 20
TIR 10.40- 12.5 (B6) - 60 -
MIR 2.09 - 2.35; (B7) - 30 =
PAN 0.52-0.90 (B8) 0.48 - 0.71 15 5
Landsat 7 ETM+ sensor provides multispectral (30m
resolution in band 1-5 and 7, and 60m resolution in band 6)
and one panchromatic band (15m resolution) image. Each
Landsat 7 ETM- scene covers a ground area of 185 km x 170
km. The earth observing instrument on Landsat 7, the ETM+,
replicates the capabilities of the highly successful Thematic
Mapper (TM) instruments on Landsats 4 and 5. The ETM+
also includes new features that make it a more versatile and
efficient instrument for regional to global environmental
change studies, land cover monitoring and assessment, and
large area mapping. Table 1 shows the detail sensor
characteristics.
3.2 Methods
Fusion of remotely sensed images obtained using different
sensors need to perform carefully. All data sets to be merged
needs accurate registration to one another and resampled to
the same pixel size.
3.2.1 Georeferencing
The scanned topographic maps at 1:50,000 scales are
geometrically rectified to the UTM projection, Zone 34 S co-
ordinates. Clarke 1880 spheroid and Cape datum were used
for georeferencing all data sets. On an average half pixel (i.e.
2m) accuracy is achieved in georefencing of topographic
maps. The geocorrected topographic maps and a few Global
Positioning System (GPS) measured ground control points
(GCPs) are used for georeferencing of SPOT 5 panchromatic
and multispectral images.
In case of SPOT 5 panchromatic image geometric correction,
64 well-distributed GCPs and 30 well-distributed check
points are selected over the entire scene of 60 x 60 km area.
RMS error for GCPs was 4.60 pixels, while for check points
RMS error was 4.30 pixels. In case of SPOT 5 multispectral
image geometric correction, 64 well-distributed GCPs and 30
check points are selected. RMS error for GCPs was 2.30
pixels, while for check points was 2.42 pixels. A nearest-
neighbour interpolation method is used for image
transformation.
Landsat 7 ETM+ panchromatic and multispectral images are
registered with SPOT 5 panchromatic geocorrected image. In
case of Landsat 7 ETM+ panchromatic image rectification,
127 GCPs are used, while in case of multispectral image, 126
GCPs are used. In case of panchromatic image rectification
RMS error was 1.79 pixel, while in case of MS image, RMS
298
error was 1.72 pixel. Again nearest-neighbour interpolation
method is used for image transformation.
The relationship between the original pixel co-ordinates (x,y)
and the transformed co-ordinates (u,v) in the new projection
is specified by a pair of mapping functions:
u = f(xy),
v = g(x,y), (1)
and by an equivalent pair of inverse functions:
x- F(u,v),
y = G(u,v). (2)
After geocorrection of all data sets, two test site common to
all images of the delta were selected for the present study.
3.2.2 Image fusion
The radiation recorded by a sensor is dependent on, among
other things, the spatial resolution of the sensor in relation to
the spatial frequency of the terrain (Wulder ef al., 2000).
Among the most important purposes of fusion of different
resolution images is the production of spatially improved
images suitable for classification. There are several
approaches to fusion of remotely sensed images. Image
fusion can be achieved at three levels, the pixel or signal
level, the feature level, and the object level (Forstner, 1992;
Pohl and van Genderen, 1998; Pohl, 1999). The signal level
fusion techniques have been most common in remote sensing
applications (Forstner, 1992).
In this study, two well-known methods based on the Brovery
transformation and standardised principal components (SPC)
transformation are applied to two pair of images acquired by
HRG sensor onboard the SPOT 5 satellite and by ETM+
sensor onboard Landsat 7 satellite. Effects of image fusion
are examined visually before classification.
3.2.3 Classification
Classification is the process of grouping of pixels or regions
of the image into classes representing different ground-cover
types. It chooses for each pixel a thematic class from a user-
defined set. Two main digital image analysis techniques are
available for the segmentation and classification of remotely-
sensed, spectral data: unsupervised and supervised. In the
unsupervised classification process, a computer algorithm
selects a sample of pixels and clusters their radiance values
Internatioi
into simil:
maximum
functions |
density c
classified ı
In the suj
image is a
types), by
properties
to the va
classificati
user-define
the classes
and superv
Three clas:
objective c
cover and
vegetation
of marsh,
identify le
interface ir
the most r
image er
enhanceme
classificatic
supervised
the most p«
sensing, in
classified
features. M
and simplic
Since. the
resolution
features, it
process rat
procedure. .
visual insp
topographic
On an avei
georefencin
best possibl
poor. In cas
accuracy \
multispectre
were treme
images as |
between 19°
In case of I
(27m) accu
ETM+ mult
achieved. T
presented in
