2

I have a panchromatic satellite image with a large coastline area and my goal is to delineate this coastline with a maximum accuracy and minimum amount of manual work. What is the best option to accomplish this?

I do not have either DEM nor other spectral bands, so these solutions do not work:

Here is a small area of the original 2-meter satellite image I have:

enter image description here

PolyGeo
  • 65,136
  • 29
  • 109
  • 338
Basile
  • 3,543
  • 3
  • 21
  • 49
  • 2
    If that's all you have then you really have a serious problem. You could give supervised classification a go but I have my doubts that it will return anything like what you need. If you can source some infra red bands your chances of automatically classifying sea/shore edge get better. Perhaps you could start with SRTM to approximate your coastline and use that to limit your processing area to save some cleaning up later. – Michael Stimson Sep 03 '17 at 23:58
  • 1
    I would explore image segmentation with Python's opencv package. Here is a blog post that will hopefully get you started: https://learndeltax.blogspot.com/2016/02/segmentation-using-cannywatershed-in.html. Python's scikit-image package is also excellent for digital image processing: http://scikit-image.org/docs/dev/auto_examples/ – Aaron Sep 04 '17 at 04:55
  • 1
    I got a decent result thresholding then grouping low values . What software do you have available? – user2856 Sep 04 '17 at 05:16
  • @Luke you name it. What would you recommend? Your thresholding seems pretty good, although there are some brighter water areas present on this image, too – Basile Sep 04 '17 at 09:32
  • @Aaron thanks for the information about those packages, especially scikit-image! Do you have any experience of converting non-georeferenced outputs of those algorithms into georeferenced ones (i.e. detected contours to shapefiles)? – Basile Sep 04 '17 at 10:42
  • As long as you know the corner coords, you can georeference the image with gdal and python: https://gis.stackexchange.com/a/116872/8104 – Aaron Sep 04 '17 at 15:15

0 Answers0