ınbul 2004 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
EIL EM-MI image are non-vegetation ones. When compared with figure 2, it As shown in figure 7, most wind erosion obstacles have their 
Is of colors can be seen that most non-vegetation regions are removed. corresponding borderline segments. Lines of non-interested 
regions such as forests can be masked out with available GIS-data, 
for example the center part of figure 7. Field boundaries also 
appear line structures, as shown on the top of figure 7. These 
edges can be easily removed by post processing according to 
NDVI and CIE L*a*b information. But boundaries of grassland 
are difficult to remove because they also appear good line 
structure, high NDVI value and positive a component of CIE 
L*a*b, for example the right side of figure 7. Another reason is 
that wind erosion obstacles are sometimes connected to grassland 
(bottom-right of figure 7). They have to be treated as potential 
wind erosion obstacles at this step. Small line segments of 
potential wind erosion obstacles are linked along-line-direction 
based on their inclination, direction and distances in between by 
perceptual grouping techniques. The linked lines will be usually 
longer than the initial ones and represent the borderline of wind 
erosion obstacles (Figure 8). 
  
  
lue from 0 
  
Fig. 6. Segmented CIR image 
  
Wy) (2) 3.2 Line extraction and linking 
Single tree and hedge are not of interest since they are nearly of 
no influence on soil science map. Wind erosion obstacles, no 
:ed below. matter tree rows or hedges, usually appear line structures, or at 
20) to red least can be treated as combination of line segments. As can be 
vith value seen from figure 6, wind erosion obstacles in the segmented 
image appear very good line structures. Canny edge extraction 
algorithm is used to extract these edges. Then the extracted edges 
are converted into line segments longer than a given length. 
3) = 
Wy, Wz ) Fig. 8. Linked line segments 
he point : 
3 value of 3.3 Line grouping and matching 
008856, Tree rows and hedges always have two borders since they have a 
certain width. So the extracted lines of two borders (sometimes 
only borderline of one side can be extracted if the other side is 
connected to grassland or of low contrast in the image) should be 
a*b. The grouped into one line segment. Corresponding line of the 
indirectly interested one of the same wind erosion obstacle can be 
IE XYZ determined by a search algorithm within a certain distance of the 
IE L*a*b cross-line-direction. DSM information is also considered during 
line grouping. The line pairs are combined together to get the 
centreline of wind erosion obstacles. In Figure 9, all line pairs are 
sitive for successfully combined together. 
) for the 
combined If no corresponding line found, the line of interest will probably 
hows the be the boundary of grassland and of course that of wind erosion 
ation. In : BABS 8 obstacles. For boundary of grassland, one side of it will be a 
> adopt a region with low NDVI value while the other side a homogeneous 
as in the Fig. 7. Extracted image lines region with high NDVI value. This information is also helpful to 
787