1

Can the ML be applied to an NDWI image? Or is it a separate means of classifying that is only applied on 2 or more bands? The problem I am actually facing is that without knowing that this could be a problem, I have already applied the NDWI and then the MLC on it. My question is now, how exactly does this change my results?

closed as unclear what you're asking by Mapperz Nov 21 '16 at 16:36

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

3

You can apply a Maxiumum Likelihood classification to a single band image. However, the results will not be very useful and could be achieved just as easily by simply reclassifying the single band into two or more classes based on the pixel value.

If you want to get a good classification from a single band, you have to look at spatial information, rather than pixel by pixel. As such, consider image segmentation as a step before the classification.

  • 1
    Alternatives to segmentation would be clustering (eg., isocluster, kmeans) or maximized local entropy. Clustering would operate on the entire image, clustering all of the values at once whereas entropy would operate on a specified local window (eg., 11x11 focal window). – Jeffrey Evans Nov 21 '16 at 18:23
  • But in my case, in which I am trying to determine Land/Water movement, applying ML to a single band image shows much better results than directly applying ML to the 5 band image. In the first case, waves are classified as water. In the second, they are classified as vegetation. Is it scientifically wrong to apply ML to a single band? – Alin Radu Nov 23 '16 at 11:15
  • 1
    There are no scientific problems with it. However, NDVI is not the best tool for delineating water. Try NIR_ratio, which is the fraction of the total reflectance of a pixel that originates from the NIR-band. It is calculated like this: NIR / (blue + green + red + NIR). A low value is indicative of water (and at times, also shadow). – Mikkel Lydholm Rasmussen Nov 23 '16 at 11:32
  • Thank you very much! The problem I am actually facing is that without knowing that this could be a problem, I have already applied the NDWI (NDVI before was a mistake) and then the MLC on it. My question is now, how exactly does this change my results? – Alin Radu Dec 9 '16 at 13:58

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