1 / 23

Sea Ice Detection in Radarsat Imagery using Statistical Distributions

Sea Ice Detection in Radarsat Imagery using Statistical Distributions. R. S. Gill. IICWG-II, Reykavik. DMI, Copenhagen. Recall:

rossa
Download Presentation

Sea Ice Detection in Radarsat Imagery using Statistical Distributions

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Sea Ice Detection in Radarsat Imagery using Statistical Distributions R. S. Gill IICWG-II, Reykavik. DMI, Copenhagen.

  2. Recall: Ice analysts often have to make ‘best guess’ of the position of the ice edge and the ice concentration when interpreting SAR data. ‘Best guess’ is often based on their experience of the region and on historical information. Accuracy of ice information has a direct impact on vessel safety !

  3. Why is life so difficult for the ice analysts and developers? -SAR signals over open water and sea ice region are ambiguous. • OPERATIONS: • It’s problem of SAR image interpretation. • Determining the ice edge and ice concentration • -in regions with low ice concentration, • -strong surface wind conditions, • -surface melting (seasonal) In the waters around Greenland ! DEVELOPMENTS: -All tools/products give results that are also ambiguous. -Even interpreting them is a problem for some ice analysts. - Manual interpretation of grey tone images helps. -PMR most reliable and simplest CURRENT STATUS: -Large gap between what the ice analysts need and what the tools/products can deliver. -Be realistic in expectations: For really ‘nasty’ images nothing will works !

  4. What is our goal ? To develop ‘easily interpretable’ tools/products that would aid ice analysts to discriminate between the different regions in a SAR image in an operational environment. CONTENT: 1. Proposed new algorithms 2. Test results 3. Conclusions

  5. A. Ice Edge detection using distributions • -Gamma pdf (undergoing testing -limiting case of k-pdf), • -k- pdf (planned for later) • -Scheme similar to CFAR method to detect icebergs. • -Show probabilities on a grey scale to discriminate between the different regions. B. Semi-automatic image classification using distribution matching • -Matching the image to a known region type • -Don’t throw the ice analysts knowledge away: use it ! • -No prior knowledge of the distribution function • -Using Kolmogorov - Smirov test (other can also be used)

  6. Classifying different region types using distributions Pearson diagram - Skewness vs. pmr Pearson diagram - kurtosis vs. pmr

  7. Computation scheme for computing distributions: Background region Test region Compute the probability distribution function, P(I), for the test region

  8. Disko Bay, 9th April 2000.

  9. PMR ‘image’. Gamma pdf ‘image’

  10. Amplitude image from Cape Farewell - 11th March 2000

  11. PMR ‘image’ Gamma pdf ‘image’

  12. Amplitude image - Disko Bay 2/4/2000.

  13. PMR ‘image’ Gamma pdf ‘image’

  14. Amplitude image - Disko Bay 24/5/2000.

  15. PMR ‘image’ Gamma pdf ‘image’

  16. Distribution matching to the ice region shown in red -based on PMR ‘image’. Distribution matching to the water region shown in blue - based on Gamma pdf ‘image’.

  17. Amplitude image from the East coast of Greenland, 22nd July 2000.

  18. PMR ‘image’ Gamma distribution ‘image’

  19. KS matching to the blue water region KS matching to the red ice region Dark values indicates maximum matching

  20. 1/10 ice, floes size 5-10 m Time when you most need help ! Amplitude image 29th June 2000.

  21. PMR ‘image’ Gamma distribution ‘image’. Not so convincing

  22. Conclusions. 1. Gamma distribution ‘images’ -Useful for determing the open water region in ice pack , near the coasts and in-land regions. -Easier to interpret than PMR -Complement the PMR ‘images’ Again like PMR, and all other texture parameters, Gamma ‘images’ are also ambiguous. All perform poorly when their help is most needed.

  23. 2. KS-Distribution matching: -Performs o.k.for the sea ice along the East and West coasts of Greenland. Shows potential for semi-automatically classifying an image. Advantage: Taps on the ice analysts experience of image interpretation. Currently 3 products operational: PMR, CFAR and now Gamma

More Related