0

I have a composite with four bands NIR, Red, Green and Blue. I would like help on which weight values to use when converting it to a single band which will then be used to classify vegetation. I am using Image Analysis in ArcGIS.

  • why not use the four bands if you want to classify vegetation types? – radouxju Aug 11 '15 at 14:54
  • I am trying to minimize the work involved in the selection of Training sites. If there are many bands then the DN values have a very wide range but if its a single band then clustering becomes easier and i might as well just reclassify – Rose Waswa Aug 12 '15 at 4:58
  • most classifiers will handle multi-band training sites. Have a look at the segmentation/classification toolbox if you use ArcGIS. Of course, if you want to use reclassify only, then you should have a single band (for vegetation/non vegetation, NDVI is then probably your best choice). But otherwise you should keep all bands. – radouxju Aug 12 '15 at 5:51
  • I have handled both NDVI and done segmentation. What i want is a new approach in doing the classification. I have small farms in my image as well as – Rose Waswa Aug 12 '15 at 5:54
  • I have handled both NDVI and done segmentation. What i want is a new approach in doing the classification. I have small scale farmlands that are mixed with other objects. The area am handling has not only different crops being planted but also different stages of growth for the same crop. Its a theory am testing on having supervised classification using one band. What i need is to identify different values for the crop and non-crop as well as other objects since they will reflect differently. It is having different bands contributing to the pixel value that will enable this. Thanks. – Rose Waswa Aug 12 '15 at 6:03
3

You may not have a background in the principles of remote sensing.

There are many existing vegetation indices, such as the most simple and common NDVI, that do what you are asking. Generally you should have an idea what index you are going to use based on the objectives of your analysis. Before you create an index, it's also best practice to convert DNs to reflectance before any subsequent image processing. Atmospheric correction is not strictly necessary if you are only using one image, but if you don't perform atmospheric correction, your results won't be comparable to scenes from other locations or dates.

  • I am not very well versed with remote sensing just under 4 years. I understand the concept of calculating NDVI. This approach will answer the vegetation aspect only. The objective is to identify different values for the crop and non-crop as well as other objects since they will reflect differently. Forgive my naivety, does the conversion of DNs to reflectance apply to a multi-band? – Rose Waswa Aug 12 '15 at 6:07
  • 1
    yes, conversion from DN to reflectance is needed for each band (And the conversion depends on the bands properties) – radouxju Aug 12 '15 at 6:09
  • The conversion still leaves me with how to get the weighted values so as to combine all the multi-bands to one single band with one value per pixel unlike several values in the multi-band. – Rose Waswa Aug 12 '15 at 12:56
  • NDVI keep being throw around like it is the entirety of vegetation remote sensing. This metric is just a univariate measure of photosynthetically active radiation and is normally used in conjunction with other spectral bands or in timeseries analysis to elucidate trend in greeness. For vegetation and landcover classification NDVI should be just one aspect of a multivarite model but not used solely in representing vegetation. It is good practice to use reflectance as it It may mitigate atmospheric attenuation, to a degree, but if analyzing a single image is not absolute necessary. – Jeffrey Evans May 24 '16 at 22:28

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.