Ontologies for the Sensor Web

1
Ontologies for the Sensor Web

• KSG/University of Manchester Workshop 29 Jan.
  2009
• Deshen Moodley
• School of Computer Science
• University of KwaZulu-Natal

2
Drivers for the Sensor Web

• Global increase and focus on natural disasters
  climate change
• More urgency to understand and monitor our
  natural environment
• Radical increase in sensor technology, large
  number of satellites producing terabytes of earth
  observation data

3
Drivers for the Sensor Web

• Realisation and commitment by governments to
  create a mechanism to share and exchange earth
  observation data
• Creation of the Group on Earth Observation (GEO)
  in 2005
• Aim of GEO is to build a Global Earth Observation
  System of Systems (GEOSS)
• Political technical barriers to overcome

4
A Global Worldwide Sensor Web

• open dynamic complex worldwide computing
  environment
• different organisations continuously deploy or
  modify geospatial data, processing and modeling
  services
• services must be discovered and combined to
  provide dynamic end user alerting and monitoring
  applications
• applications must hide the complexity of the
  infrastructure but must be easily changed to
  accommodate new services

5
Sensor Web as a platform for earth observation
research
6
Outline of talk

• Sensor Web What are the challenges?
• Benefits of agents ontologies?
• The Sensor Web Agent Platform
• ontology infrastructure
• Internal agent architecture
• abstract architecture for sensor web applications
• Case studies wildfire detection, informal
  settlement detection
• Current future work

7
Sensor Web Challenges

• Three core technical challenges?
• Publishing and discovering components
• Data fusion
• Context-based information extraction

8
Agents ontologies

• Multi-agent systems
• Publishing, discovering and invoking services
  over the Internet
• knowledge level communication between systems -gt
  eases interoperability, facilitates automatic
  machine interpretation
• Supported by ontologies
• semantic markup of service capabilities, data and
  tasks, agent communication (service invocation)
• semantic matching for matching services to tasks,
  aids in service discovery
• agent coordination, workflow specification/service
  composition
• But must integrate/leverage OGC web services
  SWE
• Long term vision
• Ontology driven information systems

9
Benefits of agents

• Communication at the knowledge level
• removes heterogeneity at the symbol, structure
  level and leaves only semantic heterogeneity
• individual agents can use different symbolic
  representations of internal models of the world
  and different implementation technologies.
• internal programs can be written by different
  people, in different languages at different times
  and for different purposes.
• the external interface, and operation of the
  agent is exposed in a formal and unambiguous way,
  based on shared ontologies

10
Benefits of agents (cont.)

• Agents incorporate pragmatics how to react to
  or act on new information
• reasoning engine for interpreting information and
  automatically responding to it
• knowledge, pragmatics
• knowledge for interpreting interactions with
  other agents
• internal reaction rules typically set out by its
  human owner/developer
• Provides a software engineering methodology for
  modeling and engineering complex systems, e.g.
  organization theory and the GAIA and Prometheus
  methodologies, obviously SWAP as well
• Identify and model logical components, interfaces
  and interactions in open systems

11
Challenges

• Representing concepts and reasoning about time
  and space - integral part of geo application
  domain
• Integration with current data models e.g. GML,
  OM and APIs, e.g. GeoAPI, OGC SWE services
• Building, maintaining, merging sharing
  ontologies
• Tools for developers to design, build and deploy
  agent applications,
• Abstract complexity of agent technology from
  end-users
• Relatively new and unknown technology must
  convince geo community that benefits are worth
  the extra development effort

12
SWAP

• Sensor Web Agent Platform driven by Computer
  Science_at_UKZN and ICT4EO_at_Meraka Institute, CSIR
• Advanced middleware that uses shared earth
  observation ontologies for integrating and
  processing earth observation data
• Consists of
• an ontology infrastructure
• agent framework for developing and deploying
  distributed applications application components

13
Propose Sensor Web Agent Platform

• Built on the MASII agent platform
• middleware for building and deploying agents
• in-house Java based research platform
• Abstract architecture
• semantic infrastructure ontologies, rules,
  inference engines
• methodology and tools for building SWAP
  applications agent roles, protocols, tools

14
Multi-Agent System MASII platform
15
Knowledge representation

• Humans store knowledge in three separate
  cognitive systems within the mind (Mennis Qian
  2000)
• the what system of knowledge operates by
  recognition, comparing evidence with a gradually
  accumulating store of known objects -gt thematic
• the where system operates primarily by direction
  perception of scenes within the environment,
  picking up invariants from the rich flow of
  sensory information -gtspatial
• the when system operates through the detection of
  change over time in both stored object and place
  knowledge, as well as sensory information -gt
  temporal
• Additional dimension of probability / certainty
• Intuition, suspicion, , fact
• Above model assigns uniform probability to all
  knowledge

16
Knowledge representation (cont.)

• Our approach to represent data
• entity/phenomenon being observed, the physical
  property (of this entity) being measured,
• the time and the space over which it is measured
  and the data structures containing the data
  values
• Inspired by NASA Sweet ontologies
• practical, engineering approach, concept space
  for Earth system science
• reuse entities from earthrealm, phenomena
  physical properties ontologies,
• e.g. air, water isa domain-entity
• temperature, pressure isa domain-property
• thus air temperature can be represented as a
  compound concept combining temperature and air,
  scalable to form other concepts e.g. water
  temperature, air pressure etc.

17
Ontologies for agents

• Conceptual level
• agent must describe its data or service it
  provides and the spatiotemporal characteristics
  of this data without the implementation details
• promotes conceptual/semantic interoperability
• promotes dynamic extraction and integration of
  higher level features from sensor data
• good conceptual description -gt increases
  possibilities for reuse

18
Ontologies for agents

• Technical level
• agents must still exchange and process data
• requires rich data types and data structures
  ranging from a single value at a specific time
  and space to multi-dimensional data over
  different spatial areas and varying time
  intervals
• communication is by message passing, message
  structure is required
• process flow or coordination between agents
• Mapping between levels
• conceptual lt-gt technical lt-gt current programming
  languages, e.g. Java/C

19
SWAP ontologies

• Space point, geometry, OGC spatial operators,
  within, intersects, overlaps (Cobra, Chen)
• Time instant, interval, within, before, after
  (OWL-time)
• Theme observesEntity e.g. earths surface,
  observesProperty, e.g. brightness temperature
  (SWEET)
• Probability Bayesian approach, incorporate
  Bayesian Networks, prior conditional
  probability (extends BayesOWL, Ding 2005)

20
SWAP Ontology infrastructure
21
SWAP Reasoning Engine

• SWAP ontologies are represented in the Web
  Ontology Language (OWL)
• Additional inference rules are added where
  necessary
• if ancedant then consequent
• SWAP uses the Jena rules engine
• reasons over both OWL ontologies as well as the
  inference rules

Agent inference engine
Spatial reasoner
Thematic reasoner
Temporal reasoner
Probability reasoner
22
Agent Architecture
External
Internal
Custom Ontologies
Shared Ontologies
OWL instance data KB
Custom Rules
Shared Rules
Jena
Reasoner
Agent execution engine
RDBMS e.g. postgres
Data Mapping API
Java
GeoAPI - OXFramework
OGC SWE Services
UI components
23
SWAP Abstract Architecture
Action
Action
Information
Coordination
Tasking
Data
24
Wildfire detection on SWAP
Fire Detection Client (UA)
Fire detection Application Agent (AA)
Contextual Algorithm (TA)
Fire Spread Modeling (MA)
Hotspot Detector (WA)
MSG Sensor Agent
Seviri Sensor
25
Wildfire detection on SWAP
26
Extending the Wildfire detection application
27
Informal settlement detection
28
Analysis of SWAP

• Practical demonstration of using agent technology
  for building Sensor Web applications
• Ontology infrastructure
• provides a methodology for building new geo
  domain ontologies -gt application driven approach
  -gt clarifies the scope of the ontology and
  intended usage -gt simplifies ontology
  construction
• contains temporal, spatial and thematic
  components for building geo applications
• Reasoning engine and internal agent architecture
• bridge between ontologies logic programming,
  rules store application knowledge and can be
  easily accessed and modified
• bridge with object oriented programming (Java )
  reuse current Geo APIs, data models, storage
  systems (spatial DBs), image processing
  algorithms
• Interoperable with OGC SWE, currently being
  deployed within the community
• Methodology for building ontology driven SW
  applications -gt more application logic is
  embedded within ontologies and rules -gt more
  dynamic, sharable, more suited to open
  environments
• Component based -gt plug play architecture -gt
  promotes reuse experimentation
• Workflows are public -gt users are able to follow
  processing steps to better understand a result

29
Current work

• Modeling wildfire spread predicting wind speed
  direction
• Using satellite images for flood detection in
  Southern Africa
• Mobile agents optimize network bandwidth
  processing resources
• Refine Uncertainty representation and reasoning

30
Future work

• Further applications needed to test whether
  adequate support is provided for SW application
  development
• Sensor tasking has been neglected, learning,
  feedback loops -gt Active Sensor Web
• Dynamic agent/process (workflow) composition,
  finding new combinations of processes and data to
  discover new information or dynamically match
  requests for information
• Modeling agent, predictions, querying models
• How does changes in the ontology/rules affect
  the agent execution engine. What changes would be
  needed in the Java code?
• Scientific workflows to allow EO scientists to
  use the Sensor Web as a tool to assemble, store,
  share modify experimental workflows

31

• Thank you