58
3 REDUCTION OF DIMENSION FOR THEMATIC MAPPER DATA
Six bands except a thermal band of TM 6 may be used
for land cover classification. For visual inter
pretation of image and histogram planes, the dimen
sion should be reduced to three modes as follows.
1) Selected three bands
A combination of band 1,4 and 5 was selected
because it showed the maximum determinant of variance
covariance matrix.
2) Principal components
First, second and third components were utilized
with the cumulative contribution of 95 %.
3) HSI elements
Hue, saturation and intensity are obtained by
a transform of six bands.
Image plane is generated by assigning red, green
and blue colors to the above three selected or
transformed components. Two dimensional histogram
plane can be generated to select two of three.
From the result of experiment, HSI elements were
the best for visual interpretation.
4 PROCEDURES OF MAN-MACHINE CLASSIFICATION FOR
LAND COVER MAPPING
Step 1:
Step 2:
Step 3:
Step 4:
Step 5:
generate an image on a graphic display with
use of HSI elements
also generate the corresponding histogram
on another graphic display
locate a map on a tablet digitizer and
establish geometric transform between
image and map,
classify the peaks portions in the histo
gram by checking the corresponding image
classify the rest with use of three planes
of image, histogram and map planes
Some areas in an image plane can be categorized
manually by using free cursor on the image or on
the digitizer. Some areas in an map can be also
assigned and classified into a certain class.
Figure 2 An example of Histogram plane
(a) Assignment of polygon in histogram plane
(b) The corresponding image file
Sympc
Space
of gee
B.A.Ncu
Moscow S
ABSTRAC1
used in
fore in
bination
the tech
into ort
ctical a
sis and
Pamiro-A
res of s
correlat
ture (ob
rizontal
cartomet
delling
ds of re
Figure 3 FRO-IMG command
Figure 2 shows a histogram plane and a map plane.
Figure 3 shows an example of assignment of polygon
in a histogram plane and the corresponding image
in red color ( original is color ).
5 ADVANTAGES OF MAN-MACHINE CLASSIFICATION AS
COMPARED WITH MAXIMUM LIKELIHOOD METHOD
Both of man-machine classification technique and the
conventional maximum likelihood method were applied
to a test area in Tokyo Bay Area with use of TM
data taken on January 23, 1985.
As compared with the result from the maximum likeli
hood method, the following advantages were pointed
out.
1) To able to obtain ground truth data in a real
time and to train them for classification in a
real time,
2) to be able to learn or consider the spectral
characteristics by intercorrelating the image and
the histogram plane,
3) to be able to set up the priority of classes
to be classified
and
4) to be able to give the known geographic infor
mation from the existing map.
Of speci
tation o
presents
solving
problems
eved to
kes use
account
of the E
interpre
zaics ar
fcy imerg;
microsca!
velopmen
cartograj
From tl
maps it :
an integ]
tain fra^
its geogr
complex i
preted ar
te integi
It is thi
graphs ai
the hiers
JIHHT , 1C
some cornu
as select
ality, si
to actual
well as t
nee. On t
sential d
ly manife
their abs
city (Neu
Modern
ly used i
can be pr
tometry,
metric an
tion of d
should oe
