Massive datasets: multi-spectral, multi-temporal, multi-resolution, multi-source

1
Discovery Net in Action Geohazard Modeling
Monitoring

• Massive datasets multi-spectral, multi-temporal,
  multi-resolution, multi-source
• Scientific Problem transfer basic information
  (data) to thematic knowledge
• Data mining solution exhaust data potential
• Configuration study remote sensing data mining
  strategy and approach
• Software development imageodesy algorithm, data
  mining algorithm, parallel association rule
  mining algorithm

• Basic mining functions
• Universal raster data input engine
• Automatic co-registration and geo-rectification
• Versatile image processing/visualisation engine
• Powerful and highly interactive clustering
  package
• Multi-data integration and analysis package

Ground Truth The link between space and
laboratory Case Study landslides in 3-Gorge
Dam/Reservoir Region in China

• Geohazard Data Mining on RSI include
• Identify Earthquake
• Identify Floods
• Identify Forest Fires
• Identify Eruption
• Identify Flora infestation
• Identify landslide

Landslide
Earthquake
Subsidence

• Wrap or deformation remote sensing images,
  provide more reliable and higher precision data
  set than those with the general methods, such as
  manual and semi-automatic methods.
• A fundamental function to automatically
  co-registrate a stack of imagery layers at high
  precision and speed.
• Raster data mining process can only proceed based
  on a accurate co-registrated data cube.
• Based on the primary control point calculated
  through the correlation algorithm, confirm the
  overlap area of the two images, divide the area
  in small grid, then find a best control point for
  each grid. At last, with the least square
  solution optimize the selected control points.
• Transformation function between the input image
  and the reference image, which will wrap an input
  image to a reference image.