PowerPointPrsentation

1
Knowledge based Earth Observation and Image
Mining Tools, Demos, Scenarios
2
Micalel Schroeder Hubert Rehrauer Dragos
Luca Marc Walessa Marco Quartulli Herbert
Daschiel Mariana Ciucu Patrick Heas Anca
Popoescu Corina Vaduva Cristian Iorga Matteo
Soccorsi Daniele Cerra Mihai Costache Houda
Chaabouni Daniela Molina
Analisa Galoppo Andrea Colapicchioni Achille
Valente Claudio Rosati Marco Quartulli Luca
Galli Marco Pastori Ines Gomez Amaia de
Miguel Gottfried Schwartz Andrea Pelizzari Martin
Boettchner Stefan Kiemle Eberhard Mickush Patrick
Harm
Prof. Inge Gavat Prof. Al. Stoichescu Herve
Turon Jose Valero Dr. Lucio Colaiaccomo Robero
Medri Carlo Zelli Sergio DElia Pier giorgio
Marchetti Michele Iapaolo
3
Klaus Seidel
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8

• Earthquake in China Damaged or Undamaged
  Bridges?

9
(No Transcript)
10
(No Transcript)
11
Example

• Interpretation of visual data
• We want to find the most probable shape ß which
  give the above image, with x the light direction

12
Example
13
Example
14
Example
15
Example

• The fidelity term penalizes scene with large
  mean squared difference from the visual data
• The prior probability takes into account if
  there is a preferred shape
• The generic view penalizes the models which
  have a larger variation with the considered
  parameters, in this case the light direction

16
Example
17

• The pyramid case

18

• The pyramid case

19

• The pyramid case

20

• Comparison and

21
(No Transcript)
22
(No Transcript)
23
Spectral Decomposition
24
Feature Extraction and Despeckling

• Optimization of the following ill-posed problem

25
Methodology Spectral Decomposition
Azimuth direction
Azimuth direction
Synthetic antenna
Synthetic antenna
Reflected signal
Transmitted signal
Transmitted signal
Target
Target
Reflected signal
Reflected signal
Range direction
Range direction
26
Methodology Spectral Decomposition Azimuth
Splitting
27
Methodology Spectral Decomposition Doppler
Centroid Estimation
28
Methodology Spectral Decomposition Doppler
Centroid Estimation
29
Methodology Spectral Decomposition Azimuth
Splitting
1
1
1
2
2
2
(b) SLC, Sub-band 1
(c) SLC, Sub-band 2
(a ) SLC, full bandwidth
Non-coherent targets
Coherent targets
1
2
30
(No Transcript)
31

• Comparison and

32
Methodology Spectral Decomposition Sub-Aperture
Decomposition
33
Methodology Spectral Decomposition Radar
Spectogram
Radar Spectogram 4-D function derived from
spectral decomposition and plotted in 2-D for
each pixel .
34
Radar Spectogram Generation
Preprocessed SLC Image
Selected target
FFT of Selected target
FFT
Matrix of center pixel of Inverse FFT of
window product
Sliding Window Operation
Radar Spectogram
35
Target Analysis with Radar Spectogram
Stable Targets
Azimuth Varient Targets
Radar Spectogram
Radar Spectogram
Optical Image
Optical Image
SAR Image
SAR Image
Range Varient Targets
Unstable Targets
Radar Spectogram
Radar Spectogram
Optical Image
Optical Image
SAR Image
SAR Image
36
Target Analysis with Radar Spectogram
Target Power line tower
Target zoomed by factor of 5
Variation of radar spectogram with change in
center pixel
37
Target Analysis with Radar Spectogram
Power line tower 1
Power line tower 2
Power line tower 3
Power line tower 4
Radar Spectogram
Radar Spectogram
Radar Spectogram
Radar Spectogram
Different Radar Spectograms for similar targets
38
Past years Meta-data based file access
The data archive
The meta - data
The data
39
Knowledge based Image Information Mining TSX
Ground Segment Systems
a

• Integrated in operational environments Image
  Information Mining (IIM) technologies for
  enhanced information content extraction from EO
  image archives. Operate the new functionalities
  in the TerraSAR-X Payload GS.
• Method Interface of DIMS and KIM systems. Extend
  the DIMS product catalogue with the semantic and
  image feature catalogues of KIM. Provide IIM
  functions.
• Applications
• concurrent queries of DIMS and KIM catalogue
• interactive selection of EO products information
  content
• IIM functions (explore, semantic annotation,
  detection-discovery, etc.)
• new generations of GS systems
• Envisaged missions
• TerraSAR-X, TanDEM-X
• SRTM
• MERIS
• GMES

DIMS
KIM
Services (SSE) Data (EOWEB) Information (KIM)
40
(No Transcript)
41
KIM CONCEPT SIMPLE
42
Today Interactive, user adapted, EO data content
access
Help to image classification
Suggest data
Mine Fields
Access to information knowledge
Knowledge share
Help to image understanding
43
KIM/KEO SYSTEM SCALABLE
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
(No Transcript)
48
Products Used In the Scenario

• Criteria for satellite imagery
• the sensor must be able to detect the phenomenon
  and changes
• the satellite must have frequent coverage of
  affected area
• the satellite program must deliver the data in
  due time.

Chalkida, 18.06.2009
49
Floods scenario
number of rows 45239 number of columns
26406 row/column spacing 1.25 m az./range
resolution 3.00 m incidence angle
33.2 scene center lat. 47.785 scene
center lon. 26.065
50
Classification and detection (red) of buildup
area and buildings including dams and bridges in
the vicinity of the river Prut. The in the upper
part is also decked. The false alarm rate is
negligible and includes the forest shadow. The
classification/detection was obtained using the
radiometric information of despeckled image and
the texture parameters estimated at resolution of
cca. 3 m.
51
For the same site the detection of water bodies
52
Example of minimization of false alarms by
detection based on fusion of texture parameters
estimated at resolutions of cca. 3 m and 6 m. The
false alarms have been produced mainly by the
forest shadow
53
Category Based Image Semantic Search Engine
Eg. Image structures recognitionclassification

• An interactive tool to help Image Analysts to
  faster and more precise explore image content,
  detect objects, patterns and structures in large
  image volumes.
• Method Integration of Support Vector Machines
  and Bayesian inference, semantic annotation based
  on learning and un-learning mechanisms. A Human
  Centered Concept.
• Applications
• Improved object detection and context
  understanding
• Recognition of smallest-scale objects
• Identification of damaged infrastructure
• Detection of changes from a single image
• Counting people and objects
• Mapping
• Humanitarian aid (refugee camps analysis, help
  demining, etc.)
• Natural hazards (earthquake, tsunami, flood,
  etc.)
• Conflicts
• GMES
• etc.
• Envisaged sensor data

Clouds
Sea
Desert
Buildings
Forest
Fields
Airports
Villages
Savanna
Ships
Traffic circles
54
CONCEPT FEATURE EXTRTACTION
Non-linear STFT
Multi band feature array
TSX scene
55
Image Semantics Category Based Learning
Semantic learning
Signal classes
Category learning
56
(No Transcript)
57
(No Transcript)
58
InSAR DEM Information Artefacts
59
(No Transcript)
60
(No Transcript)
61
(No Transcript)
62
Institute for the Protection and Security of the
Citizen (IPSC)
63
Damage assessment
64
Damage assessment