'anbul 2004
ie interpret-
he interpre-
together 45
es although
s. Some of
alysis sepa-
AR images
contents of
interpreted
e cases this
le the radar
h was con-
tected at all
ason is not
ition times.
y
resulted in
ure 3.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
À ee ar
Figure 3: (1) - optical image
(2) - Interpretation of optical image
(3) - SAR-image
(4) - Interpretation of SAR image
In addition to these interpretations the interpreters were asked
to appraise the SAR data quality for interpretation. The result
for the main categories is summarized in the following:
Roads: In open areas recognised in general well while it was
difficult in the built up areas.
Highways: Recognition and in most cases also identification
(number of lanes etc.) is possible.
Railroad: Detection and recognition was possible, but some-
times confusion with roads occurred. —
Development: Built up areas could be recognized well (1den-
tification of large buildings is possible, but small buildings
only could be recognised).
Agricultural areas: Appear in unusual colours (as compared
to optical data). A good identification or separation between
arable and grass land was impossible.
Ponds: Good identification was given in most cases but diffi-
culties with smaller ones.
Forest: Good detection and recognition, but clear identifica-
tion of forest type is not possible.
Figure 4 (see annex) shows as an example the 3 different in-
terpretation results (SAR and optical data) for the Copenha-
gen region obtained by the 3 operators.
It is remarkable that only one interpreter was able to identify
the golf course, while the other marked this area as unde-
fined. It can be seen also, that the amount of roads and the
areas identified as developed differ quite remarkable.
For this reason an investigation on the completeness and cor-
rectness of the interpreted features was performed.
3. EVALUATION OF INTERPRETATION
RESULTS
As already mentioned in section | neither ground truth nor
reference data was available to check the quality and com-
pleteness of the interpretation results. Because of the exten-
sive know-how and experience of the 3 interpreters with ae-
rial imagery and also the higher information contents, the
optical images have been used as reference. The interpreta-
tion of the SAR imagery was checked against the interpreta-
tion of the optical imagery for each interpreter individually.
The use of a single interpretation of optical imagery alone
was thought not to be sufficient, since the way an operators
address image objects is varying individually.
For each of the 4 regions and for each operator the inter-
preted SAR imagery was checked against the interpretation
of the optical data (see Figure 5 in annex).
As the interpretation was done by on-screen digitizing using
the Arc/View software, the length and area of objects could
be computed.
The analysis was done exemplarily for linear objects like
highways, roads and railways and planar objects like devel-
opment, agricultural, pond and forest areas. The SAR inter-
pretation has been compared with the optical interpretation.
For linear objects both were compared (buffered) visually
(i.e. checked if the lineaments belong to the same object) and
the length of the objects was computed for the optical and the
SAR data in the common buffer area. In addition the length
computation has been done separately for objects appearing
in either the optical data or SAR data only. A similar ap-
proach was used for planar features, which were intersected.
Buildings however were very often found at a position,
where intersection would yield wrong results, because of the
different geometric behaviour of the data sets. Therefore in-
stead of the building areas only the “building position" was
used, which means that the compliance of buildings in the
optical and SAR interpretations was checked.
For each object type the completeness and correctness of the
interpretation was computed following an approach presented
by Wiedemann (2003).
This approach is illustrated in Figure 6.
je : SAR - SAR only
Correctness - ————————
SAR
Optical — Optical onlv
Optical
~
atr rers Î
se
= interpreted in optical only (not seen in SAR)
sssssessssss interpreted in optical (sum of all)
interpreted in SAR (sum of all)
»sss«sss interpreted in SAR only (not seen in optical)
5
J Completeness =
s tere
|
Figure 6: Definition of Completeness & Correctness for lin-
ear objects .
The correctness (range 0.0 up to 1.0) is the percentage of cor-
rectly interpreted line / planar features i.e. the percentage of
the interpretation of the SAR images, which is in accordance
with the reference (interpretation of the optical data).
The completeness (range 0.0 up to 1.0) represents the per-
centage of the reference data which is explained by the inter-
preted SAR data, i.e. the percentage of the optical interpreta-
tion which could be interpreted from the SAR data. The
computation according to Figure 6 is done by using the
length in case of linear objects and the area in case of planar
object.
The results of selected object types together with the range of
the computed values for the 3 interpreters and the 4 test sites
are shown in figures 7 to 10.
