TerraSAR-X Virtual Earth Observatory

We applied the Virtual Earth Observatory principles and the knowledge-discovery and data mining framework presented above to data from the DLR TerraSAR-X archive, and demonstrated the development of a semantic catalogue and rapid mapping applications. The dataset we worked with contains 109 scenes from the TerraSAR-X archive as shown in detailed in the table bellow. The prime target types are urban areas, settlements, infrastructures (e.g., airports, ports/harbours, barrier lakes, etc.), industrial sites, and military facilities.

Type of areas

Scene location

Urban and infrastructure areas

  • Africa – 5 scenes

  • Asia – 21 scenes

  • Europe – 48 scenes

  • Middle East – 8 scenes

  • North America – 16 scenes

  • South America – 11 scenes

The following table presents the number of scenes annotated in the course of the project and the number of retrieved categories. The last column of the table gives some details about the methodology used for the annotation. The semantic categories that were identified during the semantic annotation process are organized in a two-level hierarchical structure presented graphically in the figure bellow.

 

No. of scenes /
No. of patches

No. of semantic categories

Methodology

Phase I

39 scenes

40,307 patches

336 categories

  • Support Vector Machine

  • Relevance Feedback – ranking

Phase II and after

109 scenes

110,000 patches

850 categories

  • Support Vector Machine

  • Relevance Feedback – ranking

See the deliverables of WP6 and the paper Building Virtual Earth Observatories Using Ontologies, Linked Geospatial Data and Knowledge Discovery Algorithms for more details.