matrices. Output is a maplike printout with symbols each meant
to correspond to a given type of land use.
EXIO This programme can produce a maplike printout of radiation classes
within one spectral band, the upper and lower limits of which can
be preset according to statistical information obtained by STATS.
It also allows the mapping of classified results of arithmetical
combinationssuch as interchannel radiance ratios. An object iden-
tifiable in a given channel but whose radiance range overlaps with
that of other objects in another channel can be separated first
and the corresponding multispectral information can be masked i.e.
rendered temporarily inaccessible to further classification.
This then allows separations in the other channel that was pre-
viously blurred by the overlap.
4. Procedure
Semi-automated mapping and surface assessment of selected land use cate-
gories requires that on each separate CCT the specific spectral signatures
of each category be known. These signatures can only be derived from the
radiation in the various spectral bands reflected by homogeneous training
areas the land use of which is known. It is therefore essential that
pixels belonging to such training areas can be exactly located on topo-
graphic maps or on cadastral plans, that line and element numbers of the
pixels on the CCT can be translated in terms of latitude and longitude.
4.1. Location of the test site on the CCT
In a first approximation line and element numbers amply delimiting the
area in which the test site was certainly located were roughly deduced
by mensuration on the multispectral colour composite bearing the same
identification number as the CCT. Of this area the video data were then
transscribed into ORSER format (COETRAN or SUBTRAN programmes) and a map
like printout was generated.(NMAPW programme) on which certain landmarks
outside the test site such as major highway crossings, major agglomera-
tions, forested areas with peculiar outlines, could be recognized.
On this large printout, which for reasons of economy can eventually be
produced in a reduced version (every nth pixel), the test site pixels can
be more accurately situated. Further computer work was restricted to the
test site area. Its form, rectangular on the printout, corresponds
to a parallelogram on the man. For more accurate location of pixels, map-
like printouts were generated for the channels providing the best con-
trasts.
For transitions between areas at least 3 (hor) by 2 (vert) pixels large
(170 x 160 m) which in a given spectral band both appeared as relatively
homogeneous, it was generally possible to prelocate the pixels within
100 m on the 1/25.000 NGI topographic map while field checking often
allowed an almost exact location. Intermediate pixels could then be
positioned by interpolation.
4.2. Selec
Training a
meet the f
to locate
entire sur
The minimu
Moreover t
shape in o
straddling
Spectral h
spectral b
nized refe
out deforn
Zoom Trans
between tl
field, alw
4.3. Asce:
The data «
entered ii
it was pos
spectral 1
These rad:
statistic:
As radian
and the M
sole inpu
The infor
ly by EXI!
cultivate:
well sepa
tion in b
The CANAL
The main
classific
tral band
the great