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 
Maik Netzband, Gotthard Meinel, Regin Lippold 
Institute for Ecological and Regional Development Dresden, Weberplatz 1, D-01217 Dresden, 
e-mail: maik.netzband@pop3.tu-dresden.de 
KEYWORDS: Image Classification, Human Settlement, High Resolution Satellite Imagery, IRS-1C, Data Merging, Urban and 
Regional Development Monitoring. 
The development of a hierarchical classification scheme is presented which combines spectral and morphological analysis methods. 
For an 1RS-1C merge product, a multispectral classification is performed to separate large-area landuse classes, mostly in the urban 
fringe, e.g. water, forest and agricultural areas, excavation sites, etc. Areas covered by vegetation can be separated in an exact manner 
using the vegetation index NDVI (ratio of red and near infrared channels). Concerning pure settlement areas, a combination of filter 
techniques (top-hat and bottom-hat transformations) followed by a thresholding (hysteresis-threshold) in order to identify settlement 
structures is presented. Furthermore, a classification of urban structure types (mainly residential areas) by using their morphological 
and textural characteristics is developed. For both the precise identification of urban (settlement) areas, as well as the urban structure- 
type classification, there is an urgent demand by spatial planning authorities (urban and regional planning). 
Spatial planning requires information on landuse status in ever 
shorter time intervals and higher spatial resolution. This 
demand can only be met by using new high resolution satellite 
data. The Indian IRS-1C is currently delivering data with a 5- 
metre ground resolution; other sensors (IKONOS-2, Orbview 3) 
scheduled for 1999 will provide data in the 1-metre resolution 
range. The project assesses the suitability of IRS-1C data for 
diverse planning requirements, e.g. updating of landuse plans, 
municipal survey maps, maps of urban structure types and 
biotopes, surface-sealing surveys, and working maps for 
landscape planning. It also examines the potential of IRS-1C 
data for providing the basis for general data updating at a scale 
of 1 : 25,000. This paper focuses on the use of IRS-1C data for 
automatic classification in urbanized areas and identification of 
settlement structures by combining spectral with morphological 
information contained in the panchromatic data. 
Multispectral pixel-based classification is basically possible 
with the 3-band IRS-1C LISS data but is of limited use in the 
very heterogeneous urban areas with high frequent changes of 
landuse. In the past, there were no automatic and operational 
methods available for urban settlement detection, but only 
adaptations from other application fields (like forest or 
agricultural monitoring). In addition, no attention was given to 
the morphology and spatial pattern of the detected 
objects/surfaces in conventional classification methods, 
although the spatial pattern is a key dimension of human 
settlement analysis and planning. For these reasons, it is 
necessary to develop new methods and concepts involving 
morphological and textural analysis. 
Visual interpretation of morphological grey value distributions 
in high resolution, panchromatic data leads to a differentiation 
of various objects and structures. In this process, the human 
brain analyses many different features, e.g. shape, texture, 
location and spatial interrelations of image grey values. 
Automatic calculation of these parameters is hard to reach. 
Therefore, it is necessary to develop innovative techniques or 
use existing tools for analysing the morphology of the grey 
value distribution of panchromatic images and integrate them in 
hierarchical classification schemes combining the spectral and 
the morphological information of panchromatic images. 
Due to the improved geometric resolution of new satellite-based 
sensors, the analysis of built-up or more general of settlement 
areas (depending on the different monitoring scales and on task 
definition, e.g. the analysis can involve the identification of 
single objects or whole settlement areas of towns and villages) 
is faced with new challenges, which can not be solved only by 
utilizing conventional multispectral classification. For this 
reason, several current research efforts - from different points 
of view - try to develop new strategies for very heterogeneous 
and rapidly changing urban and suburban areas. 
From the photogrammetric point of view, an overview of the 
present research activities to identify man-made objects mainly 
in airborne data is given by Grim et al. (1997). Fewer efforts are 
undertaken up till now to investigate the potential use of new 
high resolution satellite data. Presently, only IRS-1C is 
delivering high resolution data (5,8 m in the panchromatic 
modus). With this resolution, it is not possible to detect single 
objects (buildings, roads, etc.), large size buildings excluded. 
On the other hand, it has to be verified whether the IRS-1C data 
are suitable for outlining connected settlement areas, i.e. for

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.