The Belgian participation The Belgian participation in the ORFEO programmein the ORFEO programme

A bilateral enrichment

Carine Petit & Joost VandenabeeleOberpfaffenhofen – 28/03/2007

Outline

• Context of the Belgian participation in ORFEO

• Thematic section and few examples

• Methodology section and few examples

• Invitation for further collaboration

• Conclusion

Context

Belgium has a long tradition of collaboration with CNES :

– SPOT (1986) and VEGETATION (1998)– CTIV hosted by VITO in Mol has been official

distributor of VEGETATION images since 2006

Belgian participation in support programme for the use of Orfeo data

• In the framework of the Belgian participation in Pléiades programme (26 May 2004, 27 M€)

• Belgium ORFEO Budget: 1,2 MEURO– Thematic section: 0,1 MEURO– Methodology section 1,1 MEURO

• Duration: 2004-2009

ORFEO thematic priorities

• GT1 : Sea and coastlines

• GT2 : Risks and humanitarian aid

• GT3 : Cartography, Town and Country Planning

• GT4 : Geology et Geophysics

• GT5 : Hydrology

• GT6 : Forestry

• GT7 : Agriculture

Thematic section: Belgian participation

• GT1 : Sea and coastlines MUMM, VITO

• GT2 : Risks and humanitarian aid ERM

• GT3 : Cartography, Town and Country Planning

VUB, RUG, ULg, ULB, ERM, ING, VITO Meeting in Brussels – 9/5/2005

• GT6 : Forest UCL

• GT7 : Agriculture UCL, ULg, CRA

GT1 : Sea and coastlines - Bart Deronde (VITO)

• Belgian mapping expertise with HR remote sensing:– Sediments dynamics along the Belgian shoreline– Habitats on tidal sand banks– Water quality parameters (suspended sediments and chl a)– Atmospheric corrections above water bodies

• Capabilities in airborne hyperspectral RS discussed in the context of a Post-Pléiades mission

• VITO and MUMM communicated their requirements in terms of data and products to CNES

• Opens the door for possible joint projects• Waiting to work with the data now !

GT2 : Risks and humanitarian aid – Vinciane Lacroix (RMA) • Time as coordinator of the working group• A robust line detection tool (already in ORFEO toolbox)• Better connection with GMOSS by:

– Facilitating the transfer of knowledge from GMOSS

– Introducing SERTIT as an associated partner in GMOSS

• Deeper relationships with UNOSAT and CNES in the context of GMOSS

• Participation in a real-time exercise with SERTIT and CNES• CNES-RMA joint publications on SAR• Training of RMA in CNES

• New publications, FP7 projects or other, further contacts closer to the end users

• Sharing of scientific tools, methods, and data

GT7 : Agriculture – Dominique Buffet (CRA)

• Belgian administration, end-user and researcher needs in the framework of CAP

• Agri-environmental indicators from the Belgian SAGRIWATEL project• Definition of agricultural and environmental objects and of ORFEO

product sheets• Thematic study on a Walloon site

• New research opportunities using the ORFEO toolbox• Feature extraction and object recognition from imagery that would be

applied to agriculture and environmental purposes

• Valorisation in Belgian and European ongoing or future studies• New collaboration with French partners in internal or official

projects• New research orientation to feature extraction and object

recognition from imagery

Methodology section : Belgian participation

• One call for proposals for doctoral or post-doctoral research in October 2005

• Budget per project: maximum 200 000 EUR

• Number of selected proposals: 6

• Launch of the projects: March 2006

• April - June 2007: – first progress report – first meetings of the steering committees

>>> with experts from CNES and DLR

Selection of ORFEO proposals

• CHADE: Change detection for updates of vector databases through multi-level region-based classification of VHR data

(Wolff – ULB)

• GEMITOR: Multimodal Geo-referencing of 3D VHR Optical and X-Band SAR imagery (Jamar – ULg)

• EMSOR: Automatic Extraction of Man-Made Structures from VHR Optical and SAR Data for Change Detection (Acheroy – RMS)

• URMO3D: Combination of satellite-derived raster and vector data for 3D city modelling (Goossens – UGent)

• SYNOPRA: Synergy of very high resolution optical and radar data in forest mapping and inventory (De Wulf – UGent)

• ASSIMIV: Detection and description of geographic objects by means of multi-source data assimilation (Defourny – UCL)

ASSIMIV: Detection and description of geographic objects by means of multi-source data assimilation

• ORFEO theme– Use of VHR optical imagery with ancillary data

• Main objectives– Update and upgrade vector databases using a single

image and ancillary data• Study test sites

– Ardennes (Forest), Famenne (Open fields) and Montpellier (Open canopy)

• Images used and field measurements foreseen– Quickbird– DGPS survey for planimetric accuracy assessment

ASSIMIV: Detection and description of geographic objects by means of multi-source data assimilation

• Methodology key points– Use of shade and parallax model to check boundaries– Screening of artefacts of object boundaries between the

database and the image– Object-based texture analysis– Contextual classification

• Duration: May 2006 to December 2009 • Short presentation of the laboratory

– Lab of environmetrics and geomatics, Université catholique de Louvain

– Pr. Pierre Defourny : defourny@enge.ucl.ac.be

– Julien Radoux : radoux@enge.ucl.ac.be

– www.enge.ucl.ac.be

ASSIMIV: Detection and description of geographic objects by means of multi-source data assimilation

• Intermediate results (WP 100)– Trigonometric GIS model of shade and parallax shift

– Region-based detection of boundaries• Oversegmentation• Spectral characterization of remaining boundaries• Unsupervised classification based on ancillary data• Detection of inconsistencies between image and vector out

of a confidence interval

Shade

Parallax shift

ASSIMIV: Detection and description of geographic objects by means of multi-source data assimilation

• Forthcoming activities– Statistical outlier detection >>> ORFEO Toolbox

• Use of a priori information• Multimodal distribution modeling

– Object-based texture description• From pixels to objects• Feature selection

– Contextual classification• Self learning from ancillary data

ASSIMIV: Detection and description of geographic objects by means of multi-source data assimilation

• ASSIMIV as a part of ORFEO– New collaborations

• Partnership with “Maison de la télédétection” initiated for WP 200

– Thanks to ORFEO toolbox• Great opportunity to share methods• Easy access to state-of-the-art algorithms

EMSOR: Automatic Extraction of Man-Made Structures from VHR Optical and SAR Data for Change Detection

• Limited knowledge on the combined use and fusion of SAR and optical data

• Great interest of end-users for:• cartography• security (rapid mapping, damage assessment)

• Need of methods to combine SAR and optical data at a metric resolution

• “Images of the same object obtained by different sensors are different representations of the same reality” [Inglada and Giros, IEEE TGRS, Oct 2004]

• Measures exist of the similarity between images for multi-sensor image registration (medical imaging and computer vision)

• Extension of the similarity measures used for image registration to change detection

• Several “families” of similarity measures exist; among those valid for multi-sensor images, we have:– measures based on radiometry values and their probabilities– measures using only probabilities (e.g., mutual information)

EMSOR: Method is based on similarity measures

EMSOR: Similarity measures used as change detectors raise several questions

• How to deal with very different information sources?

• What is meant by “similarity” between, e.g., a radar and an optical image?

• How to estimate the reduction of similarity due to a change in the scene?

• Similarity measures identify a pixel as changed if the probability of its radiometry value with respect to that of the neighbours, in the two images, is not the same (i.e. beyond a given threshold)

Optical (1990)

Mutual information similarity image

EMSOR: A time series of optical (1990), SAR (1993) and hyperspectral (2004) images was tested

Hyperspectral (2004)

Derived change map with ground truth data in

• red: always present structures

• yellow: new ones

• Varying characteristics of the “similarity” images led to no “absolute” measure of changes

• Better results of change estimation obtained with Optical/HS images with the mutual information and the cluster reward algorithm measures

• Similarity images allowed distinction between vegetated areas and man-made structures

Comments

Acknowledgements• Thanks to M. Idrissa for the software development

• Researcher: Vito Alberga (alberga@elec.rma.ac.be)

• Signal and Image Centre, Royal Military Academy

CHADE: Change detection for updating vector databases through multi-level region-based classification of VHR data

• Use of three databases- Crop field database Ministry of Agriculture (CAP)

- 1:10000 topo maps National Geographic Institute- 1:1000 PICC database Ministry of equipment and transport

• Analysis of user needs and requirements

• Towards a method of change detection– Geocoding, orthorectification of images and co-registration with DB– Direct Image – Database change detection– Direct integration of the vector database in the segmentation process– Identification and classification of the new objects compared with the

vector database

C

Image Database Segmentation

New object detection

CHADE: Change detection for updating vector databases through multi-level region-based classification of VHR data

www.ulb.ac.be/project/orfeoSIGTel, IGEAT, Université Libre de Bruxelles

Promoter: Wolff Eléonore (ewolff@ulb.ac.be)Researcher: Carleer Alexandre (acarleer@ulb.ac.be)

IGEAT

Future tasks

- Classification for the object identification and change detection at object level

- Overcome some problems: occlusions, shadows, elevation displacement…

>>> Collaboration with ASSIMIV

GEMITOR: Multimodal Geo-referencing of 3D VHR Optical and X-Band SAR imagery

• ORFEO theme– Optic/Radar complementarity and interferometry

• Methodological challenge– To develop the ressources required in terms of

georeferencing of visible and radar images in a common reference frame

– To perform a pixel level fusion of both modalities

• Main objectives– Allow joint operational exploitation of optical and radar Very

High Resolution data using 3D information issued from SAR interferometry

• Presentation by Antonella Belmonte in the geometry session (on 28 March at 11:50)

Invitation to participate in STEREO II• New Research programme in Earth Observation

• Approved on 3 February 2006

• Duration: 2006 – 2014

• Budget: 25,85 M€

• Mid-April 2007: second call worth 1,79 M€

• Proposals to be submitted for 3 types of projects:

Thematic poles Spin-off projects >>> ORFEO Innovation projects >>> ORFEO Partnerships research-public-private

Research thematic priorities

ORFEO (2004-2009)

• GT1 : Sea and coastlines

• GT2 : Risks and humanitarian aid

• GT3 : Cartography, Town and Country Planning

• GT4 : Geology et Geophysics

• GT5 : Hydrology

• GT6 : Forestry

• GT7 : Agriculture

STEREO (2006-2014)

• Global monitoring of vegetation and evolution of terrestrial ecosystems

• Management of the local and regional environment (water, soil, forest, agriculture, coastal zones, urban areas)

• Health and humanitarian aid

• Security and risk management

STEREO II: International cooperation

: Belgian teams

STEREO BUDGET: 90 % EXTERNALBUDGET:10%

1010

: International teams

80

• International teams• Cannot coordinate a project• Must get 50 % external co-financing

• Call in 2006 : • 9,19 M€,15 projects with 21 international partners

To conclude …• Belgium offered money, time and working forces

• Belgian expertise was shared within 5 thematic groups

• 6 Belgian methodological studies will provide the ORFEO toolbox with algorithms and methods

• Belgium got data, French and German expertise, access to the ORFEO toolbox

• Common publications

• New collaborations, invitation to participate in a Belgian call for proposals

• Win-win experience for both parts

… thank you for your attention !