T
able
1. E
valuation
B
ased
on
F
uzzy
L
ogic
I
ndex
(C
lassification
S
tability
)
Class Name
Number of
Classified
Objects
Mean St. dev.
Min.
Max.
BrightGreyRoofL2
22
0.8503 0.1764 0.5165 1.000
DarkGreyRoofL2
26
0.8780 0.1846 0.5300 1.000
CeramicRoofL2
89
0.9715 0.0781 0.5660 1.000
WhiteRoofL2
45
0.9415 0.1265 0.5480 1.000
Discussion
A fuzzy reasoner,
SPOR
, was designed in C++ for
OWL
2 by
integrating PosgreSQL with a fuzzy
OWL
2 ontology. This
enabled the definition of classes with fuzzy data properties
and the expression of spatial relationships though properly
designed fuzzy object properties. The design of the reasoner
enabled handling of spatial relationships of objects belonging
to a single or multiple levels of analysis. Adoption of fuzzy
logic enabled the evaluation of the knowledge base through
stability measures. The integration with PostgreSQL allowed
computation of spatial relationships during the reasoning pro-
cess and ensured collaboration with current
GIS
and remote
sensing technologies.
Development of
GEOBIA
ontologies provides for critical re-
view and correction of the represented knowledge. It can also
be directly employed in a
GEOBIA
process to extract landscape
components. The adoption of
OWL
2 ensures the integration
of
GEOBIA
ontologies between them or with current Semantic
Web ontologies. Towards the exchange of knowledge with
other domains, it can be examined the integration of the de-
veloped ontology with top-level ontologies, such as
DOLCE
or
SWEET
(Belgiu
et al
., 2014).
To experiment with and evaluate
SPOR
, a case study regard-
ing building extraction was designed. Classification stability
results indicated that the segments were classified with high
confidence in their respected classes, since the lowest aver-
age stability was 0.8503. Accuracy assessment showed that
87 percent of the total number of buildings and 75.0 percent
of the total rooftop area was correctly classified. Most of the
omission error was due to the heterogeneity of spectral and
geometric signature of some rooftops. Most of the commis-
sion error was due to the spectral similarities of various bare
ground areas with rooftops. Elimination of such commission
error could be achieved if
DSM
data were available, such as in
Belgiu
et al
. (2014). The above observations indicate the dif-
ficulty of completely reducing the semantic gap.
Possible extensions of
SPOR
might include some of the
following. An earlier approach, developed by Hudelot
et al
.
(2008), aimed at the development of a generic spatial relations
ontology. It can be examined the extension and integration of
their approach within
SPOR
, to take advantage of the repre-
sentation of spatial relationships from a generic ontology.
Furthermore, approaches have been developed aiming to au-
tomatically create an ontology from given data by employing
machine learning techniques (Durand
et al
., 2007; Forestier
et al
., 2012; Bannour and Hudelot, 2014; Belgiu
et al
., 2014).
Such approaches might be employed for the automatic design
of the ontology hierarchy and the determination of the re-
quired features that define each class. Another consideration
could be the extension of the reasoner with additional fuzzy
membership functions and spatial properties.
Despite the rather large size of the knowledge base (37
classes, 300.000 segments) the reasoning process required
satisfactory time to complete (around 31 sec, on an Intel
i7 3770K). This addressed the issues reported by previous
studies (Arvor
et al
., 2013; Belgiu
et al
., 2014), regarding the
extended time required by current reasoners to perform the
classification process, on large knowledge bases.
Regarding the declarativeness in the representation of
knowledge with
OWL
2, the suggested formalism is being con-
sidered as satisfactory, as it gives the experts who deal with
the ontology a clear understanding of the properties, the type
of fuzzy membership functions, and their borders employed
for the definition of each class.
Conclusions and Prospects
Already, a significant amount of effort has been developed
within the
GEOBIA
community to extract semantic information
from images. The results of these efforts (knowledge bases,
employed strategies, extraction processes) today exist only
on paper or isolated and unrelated implementations. Thus,
the generic knowledge base is available only to a limited
number of individuals. Ontologies, on the other hand, have
been developed for knowledge exchange within the Semantic
Web. Therefore, such efforts provide an opportunity for the
development of a collaboration tool to allow exchanging and
enhancing of the developed ontologies for image extraction by
all
GEOBIA
community.
SPOR
will be released as Free and Open Source Software,
under the terms of
GNU GPL
ver. 3 (Free Software Foundation,
2007). The purpose is to integrate
SPOR
with other open source
GEOBIA
environments, such as
GNORASI
(Doulaverakis
et al
.,
2014). This would allow for an opportunity to take advantage
of the capabilities of an integrated environment and to en-
hance the design of such ontologies with the development of
additional tools such as an appropriate ontological editor.
Acknowledgments
The authors would like to thank the two anonymous review-
ers for their constructive comments which considerably
improved the quality of the final manuscript and the guest
editors for their excellent coordination. This research has
been co-funded by the European Union (European Social
Fund -
ESF
) and Greek national funds through the Operational
Program “Education and Lifelong Learning” of the National
Strategic Reference Framework (
NSRF
) - Research Funding
Program:
THALES
: Reinforcement of the interdisciplinary and/
or inter-institutional research and innovation.
References
Agouris, P., P. Doucette, and A. Stefanidis, 2004. Automation
and digital photogrammetric workstations,
Manual of
Photogrammetry
, Fifth edition (C. McGlone, E. Mikhail, and
J. Bethel, editors), American Society for Photogrammetry and
Remote Sensing, pp. 949–981.
Argialas, D., and C. Harlow, 1990. Computational image interpretation
models: An overview and a perspective,
Photogrammetric
Engineering & Remote Sensing
, 56(6):871–886.
Argialas, D., S. Michailidou, and A. Tzotsos, 2013. Change detection
of buildings in suburban areas from high resolution satellite data
developed through object based image analysis,
Survey Review
,
45(333):441–450.
Arvor, D., L. Durieux, S. Andrés, and M. Laporte, 2013. Advances
in geographic object-based image analysis with ontologies: A
review of main contributions and limitations from a remote
sensing perspective,
ISPRS Journal of Photogrammetry and
Remote Sensing
, 82:125–137.
Baatz, M., and A. Sch pe, 2000. Multiresolution segmentation:
An optimization approach for high quality multi-scale image
segmentation,
Angewandte Geographische Informations-
Verarbeitung XII
(J. Strobl, T. Blaschke, and G. Griesebner,
editors), Wichmann-Verlag, Heindelberg, pp. 12–23.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2015
497
