next up previous contents
Next: Learning Control of Indoor Up: Computational Information Processing Previous: Tree Level Measurements in

Urban Area Classification via Remote Sensing

Researchers: Jukka Heikkonen

Remote sensing technology offers the potential to economically gather land use/land cover information over extensive geographical areas. This technology is also fast and in principle repeatable on a relative short time scale, e.g. every few weeks. Therefore it is no wonder that remote sensing has found to be a valuable tool in the collection of statistical data concerning the land cover/land use on the earth.

The current state of the art in remote sensing is such that land cover classes can be generally inferred to accuracies of the order of 80-90% when the number of classes does not exceed 20-25. The derivation of land use classes, however, has received less attention so far and especially concerning man-made landscapes such as urban areas where pixel level description closer to the human view of the landscape would be needed in statistical and socio-economical analysis

 
Figure 4: six land cover/land use classes in the city of Lisbon derived using the developed method
Figure 4

During this project an approach for land cover/land use classification of urban areas has been developed. The approach permits using multi channel and multi temporal satellite data, and it can be easily adapted to the classification scheme at hand. The approach has been tested in the city of Lisbon, Portugal, in co-operation with the Joint Research Centre of the European Community. The experiments were based on three satellite images (two Landsat TM images taken in January and June 1991 and one ERS-1 SAR image taken in March 1992) and the CLUSTERS (Classification for Land Use Statistics: Eurostat Remote Sensing projects. Eurostat is the statistical office of the European Communities.) land cover/land use nomenclature. The experiments have shown the potentials of the method to handle large number (upto 35) of land cover/land use classes with reasonable accuracy.


next up previous contents
Next: Learning Control of Indoor Up: Computational Information Processing Previous: Tree Level Measurements in
Juha Merimaa
1/2/1998