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 
meadow meadow to farmland 
farmland Ä; farmland to meadow 
Fig. 5. Change detection analysis from 1991 to 1995. 
but differ thematically, e.g., if we use a unimodal forest 
training boundary with a new scene and we find at the same 
area meadowland also with a unimodal behaviour. To 
exclude such errors, after a change detection analysis we 
detect the training areas with changes and eliminate them 
from the database. Of particular importance is a large size for 
each training area. If an automated inside buffering reduces a 
training area to a size that is no longer useable, this area has 
to be eliminated. The size of the training set for each class in 
the classification should be at least 30 times the number of 
discriminating variables (e.g., bands) in the analysis (Swain 
and Davis, 1978). 
The next step is a classification strategy adjusted especially 
for monitoring analysis. We use common robust 
classification algorithms (e.g. Maximum Likelihood) in a 
knowledge-based hierarchical process to discriminate 
between the major landcover types (Rhein and Ehlers, 1996). 
After the first classification pass (after the first date), the 
results can be used to generate new training sets to be stored 
in our information database. 
Improvement of accuracy is the ultimate goal, which we try 
to reach through our methods. To measure the accuracy, we 
not only use standard accuracy assessment methods (e.g., 
contingency matrices). Use is also made of per-pixel values 
of the Mahalanobis distance corresponding to the highest 
membership. This technique is mentioned also at D’ Urso 
and Menenti, 1996. Overlay and mask techniques, as well as 
GIS database queries, make it possible to eliminate those 
pixels whose confidence level is lower than a predefined one. 
5.2. First Results 
Using the above techniques, a first test of our method was 
carried out to analyse changes in agriculture areas. Figure 5 
shows the change detection analysis within a farmland area. 
Changes have occurred from meadowland to farmland and 
vice versa from the year 1991 to 1995. 
First steps of an automated approach for training data 
selection within an integrated GIS and remote sensing 
environment for monitoring temporal changes are presented. 
The future activities will focus on the integration and 
automation of the classification strategies. An extension to 
multimodal classes, like urban areas, will be aimed at. 
Beyond this, a more formal statistical and quantitative 
approach concerning training classes has to be addressed. 
Anderson, J. R., Hardy, E. E., Roach, J. T., Witmer, R.E., 
1973. A land use and land cover classification system for use 
with remote sensor data. Geological Survey Professional 
Paper 964, U.S. Government Printing Office, Washington, 
D.C., 28 p. 
Argialas, D. and C. Harlow, 1990. Computational Image 
Interpretation Models: An Overview and Perspective. 
Photogramme trie Engineering & Remote Sensing, Vol. 56, 
No. 6, pp. 871-886. 
Bähr, H. and T. Vögtle (Eds.), 1991. Digitale 
Bildverarbeitung: Anwendung in Photogrammetrie, 
Kartographie und Fernerkundung. Karlsruhe, Wichmann. 
Campbell, J., 1996. Introduction to Remote Sensing. 
Guilford Press, New York. 
D’ Urso, G. and M. Menenti, 1996. Performance indicators 
for the statistical evaluation of digital image classifications. 
ISPRS Journal of Photogrammetry & Remote Sensing, 51(2), 
pp. 78-90. 
Drachenfels, O. v. (Bearb.), 1994. Kartierschlüssel für 
Biotop-Typen in Niedersachsen unter besonderer 
Berücksichtigung der nach § 28a und § 28b NNatG 
geschützten Biotope, Stand September 1994. Sonderreihe A: 
Ausgewählte Grundlagen und Beispiele für Naturschutz und 
Landschaftspflege, Heft A/4. Published by Niedersächsisches 
Landesamt für Ökologie, Hildesheim, 192 p. 
Ehlers, M., 1996. Remote Sensing and Geographie 
Information Systems: Advanced Technologies for 
Environmental Monitoring and Management. In: Singhroy, 
V.H., D.D. Nebert, and A.I. Johnson (Eds.), Proc. of ASTM 
Int'l. Symp. on Remote Sensing and GIS for Site

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.