L- 
on 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
Figure 12: Classification (roof planes) 
Fig. 11 and 12 show the results for the classification based 
on the eCognition segmentation and our segmentation re- 
spectively. The subset is also shown in Fig. 8 for com- 
parision. The roof of the surrounding hallway of building 
30.21 (upper right corner) is made of zinc, but classified as 
slate. This seems to be due to the resolution of the hyper- 
spectral data, because the width of the hallway is approxi- 
mately 2 m, thus only half the original pixel size. For the 
examples above, the hyperspectral data was resampled us- 
ing nearest-neighbour interpolation. We will address this 
point also in the next section. 
S RESULTS 
In this section we will present and discuss results of our 
approach. For this purpose we will focus on the central 
campus area (white line in Fig. 2), because for this area 
some reference data already exists, namely a database of 
buildings with information about their roof materials. 
Fig. 13 displays the result of surface material classification 
based on the segmentation by eCognition. For the classi- 
fication we used hyperspectral data resampled to 1 m us- 
ing nearest-neighbour interpolation. We furthermore used 
first and last pulse laser scanning data. First pulse data in- 
cludes more details, last pulse data already generalizes the 
result, because smaller details are not included. The shown 
roof segments represent those, which are also included in 
the last pulse data. The membership values of the classes 
gravel, stone, and slate are computed using the fuzzy or 
(max). A visual check of the results indicates that the clas- 
sification delivered reasonable results. Problems arise at 
1015 
  
  
    
vn 8, Saad | SET 
  
Figure 13: Classification (OR, eCognition) 
orco Us I ARMEN 
  
! BN WI: 
we AE ho WA M 
Figure 16: Classification (OR, IPF)