enhancement techniques can facilitate and improve photointerpretation dramatically.
2. Supervised classification requires considerable ground information for specific areas
to select representative samples for the classes, which determine the quality of the
results, It is therefore not suitable for unknown areas. The classification itself is fast
even for vast areas and the results in digital form can easily be incorporated in a data
base or in model calculations.
a
Unsupervised classification stands between supervised classification and photointerpre-
tation. The results are also available in digital form. Ground information is not a
prerequisite but the clusters are to be interpreted using more or less extensively photo -
interpretation techniques. It can be regarded as a sophisticated enhancement technique,
also displaying small classes, which may get overlooked with standard enhancement
techniques including ratios or principal component transformation .
A
The classification techniques are restricted to identification of object classes and
to a limited statistical analysis, e.g. acreage determination. Further conclusions or
deductions must be made by the analyst or additional evaluation methods e.g. model
calculations. The photointerpreter can incorporate some of these processes in his inter-
pretation procedure and is therefore in a better position to direct his efforts to the
information directly related to the task he has to solve. For some problems the best
solution would be to have the same person do the interpretation and further evaluation.
COMMON PROBLEMS
The analysis of the different procedures shows also that there are many problems common
to all techniques. The most important set of problems regards the determination of
characteristic features to separate the classes. In photointerpretation often a key is used
with qualitative descriptions of features e.g. brightness, colour, texture, shape, size,
pattern characterizing the different classes. In classification similarly quantitative
statistical values, e.g. means and covariance matrices of the brightness values in different
channels, must be determined. Experience in photointerpretation shows that some of the
features may change considerably even within a frame, e.g. systematic brightness or
color changes due to the relative position of the sun. Other features, e.g.the linear,
bending form of rivers, are similar in very distant places.
Comparable conditions are valid with classification techniques. In airborne scanner data,
the brightness change due to scan angle is very pronounced and deteriorates classification
10
acc
con
whe
In c
are
cha
sen:
to c
. but
val
ano
trul
Fur
dat
stai
Cor
the
anc
nor
wit
int
cla
So
to |
exp
als
col
the
the
To
see
=