IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002
Aster-Image,
Topsheet, GPS
data...
2 Digitized "in
S the field" and
N
3 sé
Jul. 2001
Figure 16
Roads digitized on top of a topo-sheet and on an Aster
.| image (Febr.2001; scale 1:25,000).
: s SR NET
Sep. 2002
Sep. 2002
Figure 18. Capturing Plot Boundaries in 2002 while Using an
Outdated Image (of 2000).
109. SEGMENTATION OF IMAGES BASED ON
OBJECT-ORIENTED ANALYSIS
In agricultural land use surveys, plots form the primary unique
sample units. Pixel-based image classification routines do not
consider the special linear features that relate to plot boundaries
and that are often seen on images. The plot boundaries are
special cover features that belong to the cover type:
infrastructure. In the past, only through visual interpretation,
such linear features could be considered; the quality of the
interpretations was however related to the knowledge and skills
of the interpreter.
Better tools that map the primary survey units (plots) in a fast,
standardized, and repeatable way, support survey preparation
and post-fieldwork image classification (see $7; Fig.19 and 21).
At present, a statistically highly advanced GIS tool is available
(eCognition) that 1s able to identify objects (fields), and that
segments. an image based on object boundaries (field
boundaries). Spectral noise of pixels within objects is dissolved
into the object's spectral statistics. Figure 20 shows software
settings that regulate object size, shape, permitted internal noise
(color), and boundary smoothness.
After segmentation, through classification and use of expert
knowledge (packaged into fuzzy logic relationships with other
GIS layers), objects with similar spectral characteristics can be
linked to a user defined cover class or to different classes when
the fuzzy logic relationships dictate so. The software allows,
after object classification, to merge generated object layers,
generated with different software settings, pending on the object
class under review.
200 20 01D Tan 000 1130 1300 1470 1640
339 405 .560 195 030 1102 1100 125 1830
uer eg uber n0
I EE UN Tee rada doo
E TS WITTEN TOP es ee
170 419 6588 01.6 M85 141.4 166.3 1911 2160
50 351 653 064 1255 1566 1850 2150 248.0
do 399 705 1013 1320 1626 1015 27
Figure 19. Segmentation of an Aster Image of a Humid Zone
in Ghana; 9 Bands are Used.
Due to trees in fields, plot boundaries are hardly detected; natural
boundaries of cover types are. The objects are not yet classified.