International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
changed. (See Fig. 1.) 
Besides the parcel, the pixel can also be used as the 
computing unit in T2 data. The statistic Z value of each pixel 
can be calculated based on the following equation: 
> 
7 Eu M 
ijk ic ji 
Zu = FL 
/ 
; oO 
N 
i=] icy 
Where, i is the image band number, N is the total number of 
C ix : S Vor ; 3 : 
the bands, is the specified class, "' is the pixel value in 
hide Hi. . n 
(,k) in i band. ^ "^ is the mean value of the class C in i 
band, ^ is the variance value of the class C in 1 band, and 
j.k are the column number and the raw number of the image 
respectively. 
3.4 Recognition of Changed Classes 
The changed class can be recognized through the automatic 
matching between the remotely sensed knowledge database of 
all land cover classes and the extracted statistics in that parcel. 
Multiple criterions and the Decision Tree are the effective 
methods. 
3.5 Detection and Recognition of Crossing Parcels 
In the case that the changed region in T2 data is 
corresponding to a part of a parcel or corresponding to several 
parcels in T1 data, the image segmentation method can be 
used to divide the specific region in T1 
  
Fig. 2. Updated Land Use Map 
data into several uniform parcel units, and the same method 
described above can be applied in each divided unit to fulfill 
the change detection and the class recognition. 
4. EXPERIMENTS 
Based on the method presented in this paper, the software for 
class. knowledge-oriented automatic land cover change 
detection was developed using AUTOCAD and VC++6.0, and 
the land cover maps of Shenzhen city, China were updated 
using TM 30m multi-spectral data, SPOT 10m Pan data in 
2000 and the land cover maps in 1999. Compared with the 
change detection using multi-temporal RS images, the method 
presented in this paper has the better class recognition 
accuracy up to 90%. Fig. 2 is an example of the updated land 
cover map. 
5. CONCLUSIONS 
The approach, that automatically detect the land cover 
changes in the case that time one (T1) data is existed land 
cover map and another time (T2) data is remotely sensed 
imagery is put forwarded in this paper. Experimental results 
and the actual applications show the efficiency of this method. 
It could be enriched and further improved in the later research. 
REFERENCES 
[1] Dobson, Eric L. Spatial and Temporal Autocorrelatión in 
the Analysis of Landsat Thematic Mapper Digital Satellite 
Imagery. Dissertation, University of South Carolina, 1998. 
[2] BLASCHKE, T., LANG, S., LORUP, E., STROBL, J., 
ZEIL, P. (2000): Object-oriented image processing in an 
integrated GIS/remote sensing environment and perspectives 
for environmental applications. 
[3] WU, J. (1999): Hierarchy and scaling: extrapolating 
information along a scaling ladder. In: Canadian Journal of 
Remote Sensing 25 (4):367—380. 
[4] SHEIKHOLESLAMI, G., A. ZHANG, L. BIAN (2000): A 
Multi-Resolution Content-Based Retrieval Approach for 
Geographic Images. In: GeoInformatica 3 (2): 109—139. 
1162