803 
’LING 
in the 
. for 
¡ultural 
.on based 
i for an 
ild be mainly 
or by 
After having gained an insight into PC transformations, an interpreter will 
have to define his own special sampling procedures from experience. This should 
be possible with a relatively cheap pre-processing setup using either a mini 
computer or an optical device for the linear combination of MS pictures. The 
Principal Component transformation, eventually preceded by ratio processing, 
appears to be extremely suitable as a standard pre-processing method, since, 
by means of a simple linear (orthogonal) transfornation, it results in a new 
set of pictures which are uncorrelated for the classes of interest - defined by 
the sample set. 
RECAPITULATION OP THE SELECTIVE LIMITED SAMPLING METHOD 
lly sample 
mes will 
iterpreter. 
; classes 
> PC 
’al classes, 
.ther class 
rtween A 
the other 
5 the 
lical, 
; samples 
: may, 
5 Of C 
;he 
ie will 
3 in the 
.nformation 
.ous 
;he original 
;ion 
_s can be 
3 a shadow 
-lumination 
ct PC’s,. 
;ion 
3 it can 
3S contain 
interpreter 
: be 
be evaluated 
a) Delineate area of interest (discipline) 
b) If point sampling not feasible, sample areas with a regular grid and 
perform a sub—optimal PC transformation 
c) If selective sampling possible, define classes to be separated e.g. sun 
and shadow areas, different types of crops, etc. and perform PC transform 
ation. 
d) Check all PC pictures to see if classes defined can be separated by using 
only a few PC pictures, say 3. If not possible, go to (c) and allocate a 
greater weight to those classes where separation is difficult by repeatedly 
sampling the areas containing these classes. 
If separation possible, concentrate the interpretation on the few PC 
pictures and omit all others. 
e) If a new set of MS images is obtained and the same set of classes as defined 
in (c) are to be separated, use the previously successful transformation 
matrix and go either to (d) or to (c). 
f) If experience shows that nearly all classes can be separated with the same 
transformation, standardise this transformation for the discipline concerned 
such that only the PC's of interest are obtained. Care should, however, be 
taken with regard to regional and seasonal variations. 
Note 1. This procedure is also applicable for the construction of colour 
composites in which case 3 PC’s are required. The PCT should, however, be 
optimized to perceptable colour differences rather than to grey-level 
discrimination. This can be achieved by combining the colour triangle with 
the differential colour sensitivity curve for the eye. Further elaboration 
of this idea is out of the scope of this article. 
Note 2. Vegetation classes are often separable by temporal effects. One way 
to classify these would be to optically compare two or more sets of PC 
pictures of different seasons in the flicker mode of presentation. Another 
possibility is to consider the M bands of MS records of say dates d^, dp, 
and d. as one large set of MS records with 3 M spectral bands. The sampling 
methods outlined in section 3 are applicable to this set.