image may be intepreted as tree land or grass land in 
terms of the features extracted from image (see Figure 
7 )and some rules. When the land use map is used, 
we may find that the indeterminate region is surround 
by region G, of digital land use map, so it may 
further be interpreted as a swamp meadow by using 
the knowledge; vegetation in a swamp is swamp 
meadow. 
segmentation image 
R,—tree/grass land 
Rı 
  
  
  
  
Cs 
  
Gy 
G; 
Digital land use map 
  
  
a 
Gi C; ien 
  
  
  
  
  
  
  
Synthetic land unit map 
G;—swamp 
Figure 7. An example of image interpretation 
4. EXAMPLES AND CONCLUSIONS 
Following the thought of this paper , we have 
of 
automatically converting and interpreting remotely 
developed a set computer for 
programs 
sensed imagery. These programs have been used in 
real projects including the vegetation classification of 
shuang Yang county in Jilin province of China using 
TM image,satisfactory results have been obtained. 
The TM image used in the experiment is a 512X512 
pixel image of an agricutural scene . Figure 8 is a 
segmentation image obtained by a maximum likelihood 
classifier based on ARIES image processing system 
. Because the classification is only based on spactral 
and statistical information of each pixel ,and ignore 
the the 
corresponding between is 
adjacent  pixes, 
and objects 
imterrelation between 
segments 
indeterminate. For example, in Figure 8 the willow 
882 
land may be mixed with poplar land. 
Figure 9 is a synthetic land unit map composed of the 
segmentation image and digital land use map that is 
acquired by non — remotely sensed way . We may 
obtain some relations which are difficulty found in 
remotely sensed image.In Figure 9, rivers, roads, 
resident lands etc * are provided from the land use 
map. The relation may be used‘ for image 
interpretation. 
Figure 10 is the region labeling in the complexes. It 
shows the enclosed region with the aim of forming 
correlation list, region relation list and attribute 
of the 
correlation list,land unit relatability list etc » 
list. By making use complex labeling, 
a class 
interpretation image is made. Based on the spatial 
parsing of the class interpretation image,the relation 
between willow land and irrigation land is used for the 
interpretation of willow land . As a result,the mixture 
of willow lands with popJar lands is reduced at least 
about 2594. 
5. ACKNOWLEDGEMENT 
This paper was sponsored by the National natural 
. This is 
Science Foundation of China support 
gratefully acknowledged. 
REFERENCES 
C1J Andrew K * C +, “Recognition and shape 
synthesis of 3 — D objects based on attributed 
hypergraphs” ,T—PAMI 11,1989,279— 289. 
(2) Guan Z ,“A study on the image rpresentation”, 
Wuhan Technical 
Mapping. 
University of Surreying and 
(3) Li D * ,“Thematic image interpretation based on 
spatial reasoning”, Proce « of GIS’ 92 Meeting, 
Wuhan. 
(47 V + A + Kovalevsky , “Finite topology as applied 
to image analysis”, CVGIP 46,1989,141— 161.