PSEUDOCOLOR PHOTOMAPS PRODUCTION USING NEURAL NETWORKS
Karel Charvát, Vladimir Cervenka
Geodetic and Cartographic Enterprise, Prague
Remote Sensing Centre
Introduction
Gathering of information on the land use belongs to the main
goals of remote sensing methods. This task is of special
importance in regions with complicated structural zoning, e.g.
in urban aglomerations and their surrounding. At present,
Thematic Mapper (TM) data are frequently exploited for these
purposes. Therefore, a great attention has been also paid to
the development of their interpretation. Individual methods
differ each other as for the complexity and the level of
automation. There are a simple methods based on visual
interpretation and also fully automated processes based on the
principles of artificial intelligence. Selection of a suitable
method can be made from different points of view. The most
important of them will often be the costs of processing. Lower
price is one of the main reasons for using the combinations of
visual and computer methods.
A simple method of production of 1 : 50 000 pseudocolor
photomaps was described in [1]. The principle of this methods
consisted in the production of a pseudocolor composite and its
superimposition into the line map. The process described was
divided into four steps:
a) transformation of the primary TM data into three color
components,
b) histogram equalization of separate color components,
c) refining of the pixel size,
d) color composite production from three components and its
superimposition into the map.
There will be described a new improved technology in this
paper. It applies modern principles of artificial
intelligence, especially from the branche of neural networks.
Transformation of the primary data into three color components
Usual way of color composite production consist in the
selection of three spectral bands, their conversion into the
analogue form on the film material and common projection with
red, green and blue filter by means of the multispectral
projector. However, this way is not suitable for TM data,
because only a part of the information, contained in six or
seven TM bands, can be exploited. Therefore, it seems to be
useful to transform all disponsible spectral bands into three
"synthetic" ones.
The use of the transformation called "Tasseled Cap" for this
purpose has been described in [1]. The results obtained by
this transformation was satisfactory but this method had also
some disadvantages. At first, the data had different dynamics
in the individual bands, and in some cases the part of
information had been lost.
A new transformation will be described here. It enables to
comprimate effectively the whole information contained in
original'TM data into three new synthetic channels, as well as
its decomprimation back into the original channels. This
224