I was wondering how to identify small structures (eg. beacons) in a water body? I have worked with RGB images and its almost impossible to classify waterbodies in these images.

How does the IR band help here? Will they be good enough to say that this pixel belongs to water body and this pixel doesn't?

I hope I am clear enough. Kindly ask if you need more information.

  • 3
    What type of imagery are you using for your classification? – artwork21 Jan 28 '13 at 12:57
  • Aster 5m resolution – Naresh Feb 1 '13 at 10:35
  • oops.. Aster 15m.. But now thinking rapideye 5 m – Naresh Feb 1 '13 at 11:33
  • What material are the objects made of? – Fezter Feb 5 '13 at 2:21
  • mainly metallic.. may be small boats, beacons, oilrigs, platforms, stuff like this – Naresh Feb 5 '13 at 10:12

IR is usually most useful for mapping vegetation and water, in as much as water does not reflect ir. My guess is that both your water and beacon will not reflect ir. If you have a relatively homogenous background of water you might try a segmentation approach. You will need a few pixels per feature, so if they are 20-30m structures than 5-m pixels will be pushing the upper limit of useable resolution. Do your structures have specific spectral characterisitcs or are they variable?

  • no. No spectral information I can attach. For now, I am going for 5m Rapideye and structures are 20x20 to 30x30 m. – Naresh Feb 3 '13 at 14:07

Near infrared band in satellites built for cartography have a meaninful result between water and other structures. My guess is results based on the size of the object would be less than meaningful, lots of false positives. Think flocks of white birds or any other surface or near subsurface objects as well as differences in water temperature, shallow water pans, great deal of variability, say mangroves, swamps, wetlands, oceans, ponds, lakes, waves, debris, water plants, animals, subsurface structures, micro islets, subsurface clutter.

Would be hard to pick a true signal from litter unless tuned say, for still deep water. Linear features are easier to extract or general features such as vegetation, points not so much. All depends on how accurate the result you want needs to be.


Can you assemble a vector layer from known beacons and try to do a learning process with the raster?
If you verify the signature of the raster and ir bands your looking for your classification would be very clean.
Not knowing your location. Unless your interest area is open ocean I have had success finding navigation aids gis data.
For Florida
For Hawaii and other data

  • well. The idea is good but all the beacons are unknown. May be I can visually test and make a test set. Not sure though. – Naresh Feb 8 '13 at 9:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.