Semantic search applied to Earth Observation products

Seoul, Korea - 2012.10.10
82th OGC Technical Comittee

How to search for Earth Observation imagery that contains coastal cultivated areas ?

Semantic content extraction from image is a complex and time consuming task. A simpler approach is to use the metadata footprint against exogenous data to perform image characterization.
SLACkER (SimpLe Automated Characterization of EaRth observation products) uses Global Land Cover 2000 classification to perform automatically such characterization

Keywords: semantic search, Earth Observation, EO, satellite imagery, mapshup, Land cover

Transcript

1. Semantic search applied to Earth Observation products Jerome Gasperi @ CNES | 82th OGC Technical Commitee | Seoul, Korea - October 10th, 2012
3. ? How to search for Earth Observation imagery that contains coastal cultivated areas ?
4. Orthorectified image Characterized image This is urban This is water This is forest What we got What we need
5. How to do this ?
6. Semantic content extraction from image is a complex and time consuming task
7. A different and simpler approach is to use the metadata footprint against exogenous data to perform image characterization
8. SLACkER SimpLe Automated Characterization of EaRth observation products
9. SLACkER uses Global Land Cover 2000 classification to perform automatic characterization of Earth Observation products
10. World land cover TIFF file ~ 650 Mo 1 km resolution 1 color = 1 thematic class 22 classes Global Land Cover 2000
11. Metadata Identifier Description Copyright Acquisition date Acquisition angles Thumbnail ... + Footprint SLACkER 1. Get footprint bounding box 2. Get corresponding GLC2000 area 3. Process characterization 4. Store results within database Ingestion Search
12. Tag table images within database glc2000 >> Add classification columns >> Processing 10 Crop GLC2000 raster : 24137 6169 156 119 Polygonize extracted raster Process footprint against Global Land Cover Store classification => 40.89% of Deserts (10 + 14 + 19) => 31.89% of Herbaceous (9 + 11 + 12 + 13) => 14.71% of Water (20) => 9.27% of Cultivated (15 + 16 + 17 + 18) => 3.2% of Forests (1 + 2 + 3 + 4 + 5 + 6) ( 28.65 % of 14 : Sparse Herbaceous or sparse Shrub Cover ) ( 6.1 % of 18 : Mosaic: Cropland / Shrub or Grass Cover ) ( 12.24 % of 19 : Bare Areas ) ( 31.88 % of 12 : Shrub Cover, closed-open, deciduous ) ( 2.5 % of 3 : Tree Cover, broadleaved, deciduous, open ) ( 14.71 % of 20 : Water Bodies (natural & artificial) ) ( 0.13 % of 15 : Regularly flooded Shrub and/or Herbaceous Cover ) ( 2.81 % of 16 : Cultivated and managed areas ) ( 0.69 % of 4 : Tree Cover, needle-leaved, evergreen ) ( 0.21 % of 17 : Mosaic: Cropland / Tree Cover / Other natural vegetation ) Processing result example
13. https://vimeo.com/51045597
14. What's next
15. md SLACkER classification Provide classification service through WPS
16. Segmentation+classification processing with OTB* Additionnaly, provide a "true" WPS classification service based on the OTB suite service *OTB is the Orfeo ToolBox - http://www.orfeo-toolbox.org/otb/