×

You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Title
Mapping without the sun
Author
Zhang, Jixian

221
E r „(0,0) =
/4
(8)
4. FUSION RESULT & ANALYSIS
% n >,0)+ E^Oj)*'
K E pemH (l,0) + E panH {l,l) j
00
01
10
11
00
01
02
03
10
11
12
13
20
21
22
23
30
31
32
33
2-a Pan L 2-b Pan H
Figure 2 Resampling Schematic Diagram
The energy ratio images should be converted from low
resolution to “high resolution” based on the principle that 1
pixel of low resolution is divided into N pixels of high
resolution, and every pixel value of “high resolution” equals to
the pixel value of low resolution.
3. SSVR FUSION ALGORITHM FLOW
4.1 Study Area & Data
The Study data covers Landsat7 ETM+ RS image obtained on
July 1 st , 1999, and WRS id 123/032. Because scope of
panchromatic wave band covers most part of visible light and
part of near infrared and this band’s wavelength width is similar
to that of band 2, 3, 4 of multispectral image, image covering
these three bands is merged with panchromatic image and the
spectra properties of fusion image are pretty similar to that of
origin multispectral image.
Table 1 Wavelength Region & Resolution of Selected RS
Images
RS Images
Wavelength Region
Resolution
band2: 0.525-0.605
ETM+ MS
band3: 0.630-0.690
30m
band4: 0.775-0.90
ETM+ Pan
band8: 0.52—0.90
15m
Northern suburban areas of Beijing City are taken as study
areas, and the square size is 30KM*30KM. Terrain and surface
features in this area are buildings, roads, water and farm, etc.
These objects are vary in shape and size, and RS images
correspondingly vary in spectroscopy features, so fusion result
can reflect merits and defects of algorithms.
(1) Converting DN value to radiance. The conversion formula is
4.2 Fusion Result
DN K x (CT
255
According to the selected RS images, PCA, Multiplicative,
(9) Brovey, ISVR and SSVR fusion are processed respectively and
the result show below:
where L k is radiance of band k, C k ,ax and C' k m are
radiance correction coefficients, which can be acquired from
head files, and DN k is digital number of band k.
(2) Computing energy values of remote sensing images’ bands.
The computing formula is
E xs - L i x Wj (10)
where £ is energy value of band i, f is radiance of band
i, and jy. is width of band i .Take landsat7 ETM+ data as an
example, the formula is £ = L s x (0.92 - 0.52 ) °
(3) According to formula (7), the pixel size of Pan images is
resampled from 15m x 15m to 30m><30m.
F
(4) Computing x s Li . As the energy value results of MS
E Pan L
images band 2, 3, 4 and Pan Bands are computed previously, it
is easy to get the value.
(5) Computing XSP i according to formula (6).
(6) Layerstack XSP t to make fusion effect evaluation.
1 Display effect.
Overall effect (Figure. 4), partial effect (Figure. 5).
2^ Evaluation of Fusion Effect
At all the research related to the multi-resource data fusion,
there is no unique standard to evaluate the result of data fusion,
because different data sources are introduced to the fusion. Up
to now the main evaluation standard of the fusion is the
combination of the quality analysis based on visual
interpretation and the quantity analysis of statistics.
From the view of image resolution, the resolution of all the
image after fusion is enhanced, small avenue and the board of
water are clearly than before. From the view of spectral
information, the SSVR algorithms is the best, multiplicative
algorithm is not as good as the former, and the brovey and PCA
algorithms make large spectral distortion. The SSVR algorithm
is picked for this research, since the research on land use and
land cover, the spectral information is the most important factor.
(1) Entropy: the number of entropy is directly response to
the quantity of information. The high of the number of entropy,
the more information the image after fusion carrying on.
(2) Correlative Coefficient: the parameter stands for the
relationship between the image before fusion and the image
after fusion. From the parameter we can draw the conclusion
that how many changes of spectral information could be found.
In the algorithms used in the research, DN and Radiance are
used in the image fusion (DN is used by PCA, Multiplicative
and Brovey, radiance is used in SSVR). Entropy is used to