Full text: Fusion of sensor data, knowledge sources and algorithms for extraction and classification of topographic objects

International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
seconds, the full intensity of the incoming sun radiation is 
reduced to zero. 20 seconds is equivalent to 8000 scan lines. 
This data was also used for absolute calibration and linearity 
investigations. The first step is to subtract the dark current, 
which is measured before each data take and is different for 
each pixel. Then, the following corrections are applied. 
The typical effects which disturb the image data are: 
• the odd-even effect, caused by the fact that there are two 
different amplifier circuits for the signal from the odd and 
the even CCD-elements. 
• the non-uniformity of the CCD-elements, i.e. the different 
sensitivity of the individual CCD-elements. 
• the light fall-off of the optics towards the ends of each 
• electronic instabilities due to temperature variations and the 
read-out with the correlated double sampling technique. 
The first three effects can be measured, while the fourth is partly 
The accuracy of relative calibration is in the order of one grey 
value. This calibration uses an empirical procedure, which does 
not disturb the image content (Reinartz et al., 1997). The 
absolute calibration can be carried out using the sun irradiance, 
the projection angle of the sun light on the tilted cover and the 
spectral bidirectional reflectance factor. Thereby, the calibration 
values can be determined for every single band (Schroeder et 
al., 1997; Muller et al., 1998). 
Using the same electronic gain, the difference between bands 6 
and 7 in the sun calibration data is small, in the order of 3 - 4%. 
Therefore, also the uncalibrated data can be used for examining 
the topic of this paper. 
Atmospheric correction. Signal alterations due to different 
atmospheric conditions have an effect on all signature types. 
Atmospheric models are operationally used to correct nadir 
images. The results are mainly dependent on the accuracy of 
input parameters of the assumed atmosphere. Less experience 
exists on the signal alterations due to different look-angles of 
the stereo bands. 
The correction of atmospheric attenuation for the stereo data is 
still under investigation. Due to different illumination-to-sensor 
angles for the three stereo bands, each band is corrected 
separately. The problem is the correct estimation of the aerosol 
type and the related problem of determining the aerosol 
scattering function. 
Up-to-date atmospheric correction programs like ATCOR-3 
(Richter, 1998) approach the problem by a combined terrain 
relief / atmospheric model, where the anisotropy of 
backscattering is one of the parameters to be adjusted. At the 
reporting stage of this project, no atmospheric correction was 
applied to the data. 
Terrain relief effects. In nonflat terrain, effects caused by slope 
and exposure differences superimpose the remotely sensed data 
(Schneider et al., 1992; Koch et al., 1993). The removal of such 
effects is recommended before further data processing. Due to 
the delayed delivery of the data, it was not possible to calculate 
the DEM from the stereo images in time. The results presented 
here have not been subject to such correction. 
3.2. Data preparation 
Rectification. The four bands were rectified to geographic 
coordinates (longitude and latitude) with an RMS error of about 
0. 8.pixel. In undulating terrain, the accuracy of image 
rectification depends on the accuracy of the used DEM. For the 
test site, no DEM was available due to the above mentioned 
reasons. In this case, the rectification of the stereo data was 
performed by a combined image-to-map and image-to-image 
method. Especially in hilly regions, objects were surrounded by 
bright and dark bands, which are partially due to rectification 
Band combination. For visual interpretation, two artificial 
false colour composites (FCC) were created with the band 
1. RGB = band 6 / band 7, band 6, band 7 (Figure 7) 
2. RGB = band 4, (band 7 + band 6) /2, band 1 (Figure 8) 
The first FCC was used in the anisotropy approach („anisot“) 
and the second one in the multispectral approach („MS“)- For 
the second FCC, the average value of bands 6 and 7 was used as 
„green“ band in order to eliminate most of the bidirectional 
effects. For classification also all four bands were used (called 
combined approach). 
A visual and a computer-based approach were used for data 
analysis. In the former case, the FCCs were displayed on screen, 
where the extraction of information was optimised by the 
application of different image enhancement methods. The best 
results were used for interpretation. 
Fig. 7. False colour composite used in the anisotropy approach. 
Fig. 8. False colour composite used in the multispectral 
For the computer-based classification, in addition to the FCCs 
used for visual interpretation, a classification with all four bands 
was performed. Three supervised maximum parallelepiped

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.