Automated feature extraction and visualization

1
Automated feature extraction and visualization
J.M. van der Hulst, Kapteyn Institute, Groningen

• Westerbork Synthesis Radio Telescope
• produces routinely several Gb of data every 12
  hours
• not possible anymore to inspect and correct data
  by hand
• need pattern recognition methods to excise
  interference in UV data
• need pattern recognition methods to isolate
  emission in data cubes
• noise characteristics of the data well defined
  and knows

2
Example of visibility data with interference
time
baselines
FILTERED DATA
DATA
3
Result in the image domain
FILTERED DATA
DATA
4
Feature extraction experiments on image data
cubes
Data images at n velocities representing HI
emission of objects

• many sources can be present in one field
• need to use known 3D signature to increase
  efficiency of the method
• experimented with combination of methods à
  trous wavelet, region growing, watershed
  algorithms

Example of result (white contour) on low and high
resolution image