Ontology based description of satellite imageries for application ...

The problem for this situation is that the data repository must understand the matching between dataand applications (user tasks), only then it can respond to a certain query in a proper way. Theaforementioned issues are caused by missing relationships between data domain and applicationdomain, and lacking of descriptions how entities of the data domain can be linked to certain tasks ofthe application domain.3. Related worksIn recent years, many researchers [Barb 2008, Campedel 2009, Ruan 2006, Shao 2008, Silva 2004,Tomai 2008] devote to extend image retrieval methodologies from low-level characteristics tosemantic content and relationship. They focus on describing the content of satellite imagery and theunderlying processing chains in order to retrieve information of satellite images. An adaptedontology is applied in order to describe the real world objects with human knowledge for extractingthem from satellite imageries. Ontology based approach is applied in this paper to describe satelliteimageries regarding to specific applications, for which satellite imageries can be of use, instead ofdescribing information inside the datasets itself.In [Kiemle 2002], they present a model describing earth observation data within a digital library. Themodel defines generic earth observation product classes and their relationships. Primary data aregrouped into product classes, which are described by common features like creation time, and spatiotemporalcoverage. The approach can facilitate user search for data and can be extended by definingnew definitions like product types, product component types and product parameters. It is difficult tomanage the system by mapping primary data to product groups if primary data belongs to more oneproduct group.A further work [Podwyszynski 2009] presents an approach, on that satellite imageries are describedbased on properties of sensor and the related applications. The system is divided into two domains,the application domain and data domain; the application domain describes theapplication/phenomenon users are working with. The data domain describes the properties ofsatellite imageries. The two domains are related with each other through a measurement component.Users are able to search data based on applications they are interest in without any knowledge of lowlevel data characteristics. However, the system only describes properties of each image such asspatial resolution, processing level instead of describing the properties of sensors. It will be timeconsuming and errors prone for data providers especially when data is updated or reorganized and itis also difficult to extend when new sensors are built.4. MethodThe ontology-based approach is applied to describe satellite sensors and user requirement incircumstances of earth observation and the relationship between them. The concept of ontology canbe expressed as an encoding of a specific domain in order to share knowledge and understandingCopyright 2011 Shaker Verlag Aachen, ISBN: 978-3-8440-0451-9

epresented via the most common terminologies and ideas that are acceptable among differentspecific communities. In one domain, the ontology defines a set of representational terms which arecalled concepts; it also concludes constraints and rules to preserve the consistency of domain. A wellknown definition is “an ontology is a formal explicit specification of a shared conceptualization”[Gruber 1993]WavelengthSensorhasTimeresolutionCan observeReal worldobjectsProduct ofSpatialresolutionFocus onImageriesProvide toFigure 1Conceptual frame workUserThe objective of this approach is to approximate satellite data to the user requirements (which aremapped in the application domain). An ontology is built to describe sensor properties and userrequirements in a machine readable way. A reasoner can be used in order to return the appropriateresults to user queries based on the definitions in the ontology. Ontology is built based on OntologyWeb Language (OWL) .The approach is depicted in figure 1.In this approach, sensors are described by properties, such as type (optical or microwave),wavelength (visible, near infrared etc.), temporal resolution (high-, middle- and low temporalresolution), and spatial resolution (high-, middle- and low spatial resolution). These descriptionswere collected from [Satellite Imaging Corporation]. Potential applications of satellite imageriesrelated to specific sensor properties are shown in table 1 [NASA, Navalgund 2007, Wang 2010].Each sensor type is defined as a class, in which satellite images are individuals (e.g. Landsat,Quickbird, TerraSAR-X etc.). Each image has properties related to subclasses of three attributiveclasses (WaveLength, TemporalResolution and SpatialResolution). These properties arehasWaveLength, hasTemporalResolution and hasSpatialResolution.Copyright 2011 Shaker Verlag Aachen, ISBN: 978-3-8440-0451-9

Channel/WavelengthSpatialResolution(m)TemporalResolution(day)Swath (km)ApplicationOpticalVNIR, SWIR 20 – 30 15 – 25 100 – 200 Vegetation (forestry,agriculture),coastal zone and inland waters,soil,geology, Multiple cropforecastingVNIR, SWIR 2.5 -10< 1VNIR, MIR,TIRImaging stereopairsRadarL (~23cm)C (~3 – 7 cm)90250 – 10001000 - 4000Urban applicationDaily ~1000 Ocean color, Sea surfacetemperature Regional vegetationmonitoring; atmospheric, cloude(weather), parameter retrieval2.5 – 15 ~100 Elevation mapping, 3-Dmodelling3 – 100 30 - 35 Water mask, flood, cropmonitoring, forestry, surfaceland deformationTable 1.Relationship between specific satellite imagery characteristics and applicationsFurthermore, Satellite images also have properties such as processing level, spatial coverage andacquisition date (Fig 2.). Data domain is organized by classes, subclasses and restrictions in order topreserve the consistency of domain, expressed as the follows.• Quickbird and Landsat belong to Optical class.• Optical and Microwave type must have spatial-, temporal resolution and wavelengthproperties.• Each observation only has exactly one spatial-, temporal resolution. EtcObjects that can be observed using satellite images are also defined by classes and subclasses (Figure3.) and they are described regarding to user requirements, for example forest belongs to land usegroup as part of the application domain. In this case, the domain was defined by two main classes i.e.Natural and man-made classes (Figure 3.). They have many subclasses regarding to possibleapplications and real world objects. The hierarchy of classes presents relationships between observedobject and possible applications of user interest.Copyright 2011 Shaker Verlag Aachen, ISBN: 978-3-8440-0451-9

Figure 2.Classes hierarchy of Data DomainFigure 3.Classes hierarchy of Application domainThe relationship between data domain and application domain is defined by rules. These rules definewhich objects can be observed by which datasets through properties of the observation. For example:If image type is opticalWavelength between 0.45 to 0.52 (band Blue)Can observe coastal, soil, vegetation differentiation, and deciduous/coniferousdifferentiationCopyright 2011 Shaker Verlag Aachen, ISBN: 978-3-8440-0451-9

These rules are easy to extend whenever either new applications or new available sensors aredeveloped. While this approach applied, users do not need to search a certain sensor but they only dowith their application and the reasoner will provide a list of datasets from different sensor that aresuitable for their tasks.5. Conclusion and future worksThis approach applies an ontology in order to describe satellite sensor properties and userrequirements in a machine readable way. The ontology includes the rules that define the relationshipsbetween sensor properties and user requirements. This approach facilitates users search for satelliteimages and is easy to extend when new sensors are made or new user requirements using satelliteimage to observe real world objects. Extending of terms for application domain will be done to applyfor different disciplines and at present, image processing methods are not included in this approach.ReferenceBarb, A. S., et al. (2008). "Ontology Driven Content Mining for Semantic Queries in Satellite ImageDatabases."Campedel, M. (2009). "Knowledge Representation for Satellite Imagery."Chang, C., et al. (1997). "Titan: a High-Performance Remote-sensing Database."Datcu, M., et al. (2003). "Information Mining in Remote Sensing Image Archives: SystemConcepts."Gruber, T. R. (1993). "Toward Principles for the Design of Ontologies Used for KnowledgeSharing." International Journal Human-Computer Studies 43: 907-928.Kiemle, S. (2002). "From Digital Archive to Digital Library – A Middleware for Earth-ObservationData Management." Lecture Notes in Computer Science: 230-237.NASA "Remote Sensing Tutorial." http://rst.gsfc.nasa.gov/.Navalgund, R. R., et al. (2007). "Remote sensing applications: An overview." Current science Vol93(12).Podwyszynski, M. (2009). "Knowledge-based search for Earth Observation products." Master thesisof University of Passau Department of Computer Science and Mathematics.Ruan, N., et al. (2006). "Semantic-Based Image Retrieval in Remote Sensing Archive: An OntologyApproach."Satellite Imaging Corporation "Satellite Sensors." http://satimagingcorp.com/satellite-sensors.html.Shao, Z., et al. (2008). "Building Semantic Ontology Databases Based on Remote Sensing Images."Silva, M. P. d. S., et al. (2004). "Remote Sensing Image Mining Using Ontologies."Tomai, E., et al. (2008). "Ontology-Based Land Degradation Assessment from Satellite Images."Wang, J., et al. (2010). “Review of Remote Sensing Use in Forest Health”, The Open GeorgraphyJournal, 3, 28-42Copyright 2011 Shaker Verlag Aachen, ISBN: 978-3-8440-0451-9