sment 
  
  
ESSES 
els in an 
types by 
sification 
ve been 
etch was 
ch where 
orm with 
data in 
  
Thereafter. haze correction was done. i.e. 
xssArc 77 X883 - 19 ; (x3-$-19) 
Eventually, we ended up with three radiometncally 
corrected images: xsslrc, xss2ro and xss3rc. These 3 
bands were used to create the colour-composite image 
for conventional classification purpose but after 
geometric correction. 
3.1.1.2 Geometric correction. 
a) Panoramic distortion is negligible because the ratio of 
swath end element to nadir element is equal to 1.0013. 
b) Earth rotation is considerable because the ratio of the 
shift across the swath to the swath width is equal to 
0.050. 
3.1.2 Train sampling 
Eight classes were chosen to cover different land use to 
this area, as shown in table 2. Therefore, by choosing 
representative or prototype pixels from each of the 
desired sets of classes, these pixels are said to form 
training data. Training sets were established using the 
established (existing) ground truth map from a colour 
composite image on the ILWIS software. The training 
samples for a given class should lie in a common region 
enclosed by a boarder line or an edge. Unfortunately, 
our area of interest did not have clear defined 
boundaries to discriminate properly between different 
land covers due to composition of parcels of variable 
uses and outstanding height difference. Relief 
displacement has blurred the image to the extent that 
even edge enhancement does not help much to 
distinguish the parcels. 
  
  
  
  
  
  
Class Name Class Number 
Forest 1 
Orchard 2 
Urban Area 3 
Wasteland 4 
Wasteland 5 
Horticulture 6 
Grass Field 7 
Shrub 8 
fable2- Spectral classes of data 
3.1.3 Error analysis 
At the completion of conventional classification it is 
necessary to assess the accuracy of the results obtained. 
This was done by selecting a sample of pixels(test 
sampling) from remaining parcels and checking their 
labels against classes determined from training sampling, 
or crossing the two together to determine the accuracy. 
These results are expressed in a tabular form, often 
referred to as a confusion or error matrix. 
Confusion matrix is shown in figure 2. 
537 
  
  
  
  
  
  
1 2 3 4 5 6 7 8 unci ACC 
! 3 9 f 9 0 9 D 1 8 0.97 
2182 0 0 0 0 0 0 0 0 0.00 
340 0 68 1 7 2 0 0 0 0.87 
4 | 51 0 0 0 0 0 0 30 0 0.00 
510 0 9 3 1 0 25 0 0 0.03 
610 1 0 4 0 0 0 0 0 0.00 
710 0 0 0 8 0 9 0 0 0.00 
8 | 16 0 0 0 0 0 0 1 Ü 0.06 
REL | 0.18 0 0.88 0 0.06 0 0 0.03 
average accuracy x 24.08 * 
average reliability - 14.42 % 
overall accuracy m 29.82 % 
figure2- Confusion matrix of Maximum 
Likelihood classifier 
3.2 Creation of DEM and its applications 
In satellite images there is some amount of relief 
displacement and because of very high flying height 
and moving projection center, relief displacement across 
image lines is assumed to be very small and negligible 
but for pixels along the lines is not too small and it can 
be computed and removed. 
In order to give corrections to all the pixels, the image 
observations are modified by introducing artificial errors 
(corresponding to relief displacement), errors that will be 
corrected while later resampling the whole image. This is 
the same as if the observations were made on a plane or 
flat ground. Therefore we need a DEM of area to remove 
relief displacement and obviously this DEM should be 
georeferenced, so we only need an affine transformation 
to remove relief displacement: 
inb si a en T inser 
Having DEM , with help of some software we can first 
match our image to the DEM and then remove 
displacements of all pixels. The program CTPTODTM 
matches image to DEM and corrects the control points 
for relief and its usage is: 
CTPTODTM xtie outcoe pat dtm 5.6 -351 150 
xtie: Name of our control points file 
5.6: Looking angle. 
-351: Beginning column. 
150: Height reference. 
The other program which name is GCD removes relief 
displacements. The DEM should be georeferenced and 
we match the images to the DEM with control points; It 
means that we have matched the image to the ground 
and after this step the program will read heights of 
points from DEM and it applies relief displacement for 
all pixels. We will also have a list of control points with 
their heights and their displacements. It should also be 
noted that relief displacement should be output driven. 
Another application of DEM is to use it in creation of 
orthophoto and to do so we should patch the DEM with 
the program DTMPAT. 
To create a DEM one can use SPOT PAN images (left 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996