BUILDING A SECURE GEOSPATIAL SEMANTIC WEB

Raytheon Corporation and the University of Texas at Dallas

ABSTRACT
Semantic web is a collection of technologies that enable machine understandable web pages. In recent
years there has been a lot of research on the semantic web. More recently there is some work on geos-
patial semantic web and secure semantic web. Our collaborative research attempts to integrate both
geospatial semantic web and secure semantic web to develop secure geospatial semantic web.

1. INTRODUCTION
Geospatial semantic web integrates semantic web technologies with geospatial technologies and securi-
ty technologies. Tim, Berners Lee proposed the semantic web in order to create machine understandable
web pages. A semantic web can be thought of as a web that is highly intelligent and sophisticated so
that one needs little or no human intervention to carry out tasks such as scheduling appointments, coor-
dinating activities, searching for complex documents as well as integrating disparate databases and
information systems. Geospatial data emanates from numerous devices at multiple sites. Such data is
complex and heterogeneous in nature. In order to integrate heterogeneous geospatial data sources se-
curely, we need to develop semantic web technologies that handle geospatial data.
In this paper we describe the collaborative research between Raytheon and the University of Texas at
Dallas on building a secure geospatial semantic web with crime analysis as our application area. We
have focused on three major aspects: geospatial semantic web; geospatial data mining and security.
The organization of this paper is as follows. Our example application is discussed in section 2. Our
research on geospatial semantic web as well as the DAGIS system we have developed is discussed in
Section 3. Our research on geospatial data mining is discussed in section 4. Security considerations
including secure interoperability of geospatial data is discussed in section 5. The paper is concluded in
section 6.

2. CRIME ANALYSIS APPLICATION

We are particularly interested in Policy Blotter Crime Analysis. Police Blotter is the daily written
record of events (as arrests) in a police station which is released by every police station. These records
are available publicly on the web which provides us wealth of information for analyzing the crime
patterns across multiple jurisdictions. The Police Blotters are available to public or between police
departments are generated from legacy systems and may also be published as web documents.
There are major challenges that a police officer would face when he wants to analyze different police
blotters to study a pattern (e.g., a spatial-temporal activity pattern) or trail of events. There is no way a
police officer can pose a query where query will be handled by considering more than one distributed
police blotters on the fly. With the advance of Web 2.0 , there are some mashups of Google Maps with
police blotters of some counties. There is not a cohesive tool for the police officer to view the blotters
from different counties, interact and visualize the trail of crimes and generate analysis reports. The
Blotters can currently searched only by keyword through current tools and does not allow conceptual
search, and fails to identify spatial – temporal patterns and connect various dots/pieces. Therefore, we
need a tool that will integrate distributed multiple police blotters, extract semantic information from a
police blotter and provide seamless framework for queries with multiple granularities. Our research is
developing such a tool based on geospatial semantic web, data mining and security technologies.
3. GEOSPATIAL SEMANTIC WEB

1
 2 Raytheon/UTDallas

Secure geospatial semantic web is an integration of secure semantic web and geospatial semantic web.
Follow along the vision of Tim Bernes Lee for the semantic web, we have defined a layered architec-
ture for a geospatial semantic web. At the bottom layer are the protocols for communication. Next we
have the GML (Geography Markup Language) and GML schemas layer. We have developed GRDF
(Geospatial RDF) to specify the semantics and the GRDF layer lies on top of the GML layer. On top
of GRDF we have developed geospatial ontologies and query facilities. Below we give an example of
GRDF.

<owl:Class rdf:id=http://127.0.0.1/grdf_voc#City />

<rdfs:subClassOf rdf:resource="http://127.0.0.1/grdf_voc#tplace"/>
<rdf:Property rdf:about="http:// 127.0.0.1/grdf_voc#boundary">
<rdfs:domain rdf:resource="http://127.0.0.1/grdf_voc#
CRS_EPSG:6.6:4326/>
</rdf:Property>
</rdfs:Class>

In this example we have defined a City class (or concept in ontology parlance), which has a property
that identifies the boundary extent of a particular city. The City class is also a subclass of the Place
class and as a result inherits the latter class’s properties.
Web Services and the standards provided by OGC (Open Geospatial Consortium) define our approach
for a geospatial semantic web. Client queries the Service Requestor Web Service, which handles the
GIS Application. The Service Requestor then discovers the required Service Provider through the
Service Registry or the Match Maker. The Service Registry selects the Service provider which has
already registered with this registry. Service Provider can now bind with the Service Requestor to
fulfill the service requestor. The underlying protocol stack is shown in the figure. By this way two
different GIS application with different heterogeneities can interoperate with each other using Web
Services.

Service 1. Register/ Matchmaker

Provider-1 Advertise OWL-S MX
…
…
Service 3. Service Discovery,
Provider - n Service Enactment Reasoner/
Matching Engine
(GRDF)
DAGIS DAGIS
Interface Agent

2. Query

Figure 1. DAGIS System

We have developed a system called DAGIS (Discovery of Annotated Geospatial Information Services) that
reasons with the ontologies and answers queries (see Figure 1). It is a framework which provides a
methodology to realize the semantic interoperability both at the geospatial data encoding level and also
for the service framework. DAGIS is an integrated platform that provides the mechanism and architec-
ture for building geospatial data exchange interfaces using the OWL-S Service ontology. Coupled with
the geospatial domain specific ontology for automatic discovery, dynamic composition and invocation
of services, DAGIS is a one-stop platform to fetch and integrate geospatial data. The data encoding is
 Secure Geospatial Semantic Web 3

in GRDF and provides the ability to reason about the payload data by the DAGIS or client agents to
provide intelligent inferences. DAGIS at the Service level and GRDF at the data encoding layer pro-
vide a complete unified model for realizing the vision of geospatial semantic web. The architecture
also enhances the query response for the client queries posed to DAGIS interface.

4. GEOSPATIAL DATA MINING

Geospatial data mining will have applications in crime analysis as well as in developing and refining
ontologies. We are utilizing a multi-step approach for geospatial data mining for crime analysis,
border control as well as for suspicious activity detection. In this section we will describe our initial
research on developing data mining techniques for the classification of remote sensing data. Such
classifications can be used to determine whether there are suspicious activities in particular regions.
Land cover information can be derived from various remote sensing systems, such as images from
Landsat 7 ETM+, SPOT HRV/HRVIR, Terra ASTER and AVIRIS. The remote sensing images can
have different spatial resolutions and spectral resolutions. Classification on the pixel level cannot
reveal semantic concepts at higher levels, and the semantic concepts at high levels can be crucial for
security protection, environment evaluation and urban open space research. For instance, if a pixel or
a few neighboring pixels are classified as water body, the location can be a pool in a residential area,
a pond in an urban area, or a lake in a park or open rural area. Similarly, a group of buildings can be
for public service in an urban area, for residential purpose in a residential area, or for highly confi-
dential military use in a desert. It is of great need to develop high level concepts and distinguish
them so that the semantic meanings of pixel classes become clear and accesses to some confidential
concepts become controllable for security consideration.

Figure 2. Geospatial Data Mining

Our proposed approach (illustrated in Figure 2) classifies data combined from different resolutions
and forms high level concepts by grouping and re-evaluating classes of pixels. The classification is
performed by using support vector machine (SVM) classifiers (see Table 1 & 2), which have been
successfully demonstrated to outperform Maximum Likelihood (ML) and artificial neural network
(ANN) classifiers.
 4 Raytheon/UTDallas

To generate high level concepts for a group of neighboring pixels, we will exploit ontologies. An
ontology is a collection of concepts and their inter-relationships that collectively provide an abstract
view of an application domain. We will develop domain-dependent ontologies as they provide for
the specification of fine grained concepts while generic ontologies provide concepts in coarser grain.

5. SECURE GEOSPATIAL SEMANTIC WEB

While organizational resources can be protected with a semantic access control system, geospatial data
protection in a distributed environment can present many challenges beyond providing or denying
access. Geospatial data is unique in that the same piece of data has varying level of granularity depend-
ing on the context. For instance, raster images could be processed in different resolutions, scale and
accuracy. Even vector data is available at differing scales depending on the particular data collection
agency. For this reason, when data from multiple agencies are integrated, access control of the aggre-
gated geospatial data becomes a potential security trap. For instance, if a multi-layered view is pre-
sented to users with each layer belonging to a different source provider, it is not clear how a user from
one particular agency will see the aggregated view. The special traits make it infeasible for generic
semantic access control such as the one we have developed for DAGIS to govern over geospatial data.
In our research, we have defined two types of constructs so far. First type provides alternative abstract
elements for vector data and the second type constitute ontology for subject and action roles. Subjects
are classified based on their functional criteria. Currently defined top level classes for various catego-
ries of subjects are ‘Administrator’, ‘GISAdmin’, ‘SystemAdmin’, ‘Manager’, ‘Regular Professional’,
‘Facility Personnel’, and ‘Guest’. The actions defined so far are ‘Read’, ‘Write’, ‘Save’, and ‘Ex-
ecute’. In an ideal circumstance, all parties in a distributed system have an agreed-on set of measures
to combine their policies or resolve them in case of a failure. However, the pre-arrangement is not
always possible and in such cases, our security constructs would allow a semantic access control
processor to interpret the role and action definitions and combine the corresponding policies. Our
research is focusing on policy integration algorithms. We have developed a technology for developing
partial responses at the same time ensuring security. We have defined our ontology using SWOOP [1]
ontology editor. A snapshot of the ontology hierarchy is given below:
 Secure Geospatial Semantic Web 5

6. SUMMARY AND DIRECTIONS

In this paper we have discussed our initial research on developing a secure geospatial semantic web.
Our major contributions so far include the following:
 Development of Geospatial RDF for specifying geospatial semantics and ontologies
 Geospatial data mining for classification of remote sensing data
 Policy integrator for geospatial data interoperability
Our goal is to apply the technology for security applications including crime analysis and border con-
trol. While we continue with our collaborative research on building a secure geospatial semantic web,
we will also enhance the DAGIS system into a fully functional prototype that will answer complex
queries and help in decision making. In particular, our prototype will include the following compo-
nents:
 Semantic Search Browser for Police Blotters
 Tools for Generating Crime Analysis Concepts from Blotters
 Map based Visualizing Tools and Semantic Dashboard