![Results by Area [%]](https://image.slidesharecdn.com/ecognitionusersummit2009cstorchgafemlc-091112045111-phpapp01/85/E-Cognition-User-Summit2009-C-Storch-Gaf-Emlc-5-320.jpg)
![Results by Area [%] Results by Area [%]](https://image.slidesharecdn.com/ecognitionusersummit2009cstorchgafemlc-091112045111-phpapp01/85/E-Cognition-User-Summit2009-C-Storch-Gaf-Emlc-6-320.jpg)
E Cognition User Summit2009 C Storch Gaf Emlc
- 1. Euro-Maps Land Cover The Definiens Enterprise Image Intelligence Suite for operational landcover map production
- 2. Purpose Purpose Creation of an up-to-date landcover map of Germany for: Tele-communication network planning Environmental mapping / monitoring Risk management Regional planning
- 3. Product Specification GAF AG Coverage Germany Thematic Accuracy ≥ 95% per class Class Number 22 Positional Accuracy CE90 15m Ground Resolution 25m Base Satellite Data IRS-P6 LISS-III Minimum Object Size 0,25ha (2x2 pixel) Acquisition Year 2008-2006 Minimum Object Width 50m (2 pixel) Format Raster
- 4. Legend / Workflow Very Dense Urban Areas Dense Urban Areas Low-Density Urban Areas Very Low-Density Urban Areas Big buildings Impervious Surface Agriculture (Open) Water Shrub / Shrub-like Vegetation Wetland Vineyard Orchard Hop Rocks Excavated Material / Landfill Mining Fields Sand Bridges Greenhouses Coniferous Forest Mixed Forest Deciduous Forest Legend automatic image interpretation automatic
- 5. Results by Area [%]
- 6. Results by Area [%] Results by Area [%]
- 7. Data Specification IRS P6 LISS-III 43 scenes 23.5m spatial resolution 5 spectral bands: green, red, nir, swir + syn blue Acquired mainly in 2007 and 2008 (May – Sep) IRS-P6 LISS-IV Mono/LISS-III Merge 220 scenes 5m spatial resolution 4 spectral bands Reference dates: 2005-2007 (Mar – Oct) Data Specification
- 8. Data Preparation 220 scenes available Original size ~ 1.5GB GAF AG After optimization: 455 images (subscenes) < 500 MB 145 GB / 360.000 km² to process
- 9. Classification Input Additional Input Data Urban Area Mask Metadata containing an image identifier and the acquisition date The vector file containing the reference areas for each scene
- 10. Why eCognition? Why eCognition? For subsequent mapping, the main focus was on the delineation of the basic landcover types results should be as homogeneous as possible Definiens Developer provides very efficient workspace automation functions Definiens Developer enables a very high degree of automation through the reuse of rulesets GAF AG
- 11. Project Creation GAF AG
- 12. Ruleset Structure GAF AG
- 13. Subset Selection / Segmentation
- 14. Classification approach 2 main steps: Initial classification: use of only few features (ratios) for a robust but incomplete base classification Computation of statistical parameters for each class Final classification: Use of absolute values in the final classification parameter sets GAF AG
- 16. Results Results
- 17. Classification Results GAF AG
- 18. Conclusion GAF AG Definiens Developer enables easy data management & organization Workspace automation & ruleset recycling is very effective Computational effort was immense Input of smaller subscenes
- 19. Thanks for your attention !