So I have NDVI raster files for about 1000 acres of vineyards we manage. What I have done is created a model that takes the NDVi raster, reclassifies it to isolate out the areas of high vigor, then runs an Iso Unsupervised Classification, then generates a polygon (shown below). What I need to do is filter out the "noise" (aka the weeds and grass with very high NDVI values that my model couldn't filter out). enter image description here

The 3 rows on the left are polygons showing our producing vines, gaps in between represent dead / missing bines. The issue is the giant green globs in-between each row; which is essentially tall, healthy grasses etc due to someone not mowing before the drone captured this imagery.

Other then manually going in and selecting the grass polygons in ArcMaps and deleting them, is there a tool I can use to automate this process? They unfortunately do not have a distinguishing attribute allowing me to select them.


There are many different ways to go here, both in the vector and raster domain. My initial thought is to improve your classification accuracy in order to avoid the noisy polygons to begin with. I would avoid the unsupervised isodata classification algorithm in this case. NDVI alone is not sufficient to distinguish between classes of vegetation. Rather, you should incorporate vegetation indices such as NDVI along with your spectral bands (R, G, B, NIR). It would also be beneficial to incorporate texture metrics for an analysis such as this. Additionally, an object oriented segmentation approach using ArcMap's meanshift algorithm would be useful.

If you do not have the option to go back and do a proper classification, you may be able to clean up the polygons by using the following approach:

  1. Buffer the polygons by X distance (maybe 5m to begin with).
  2. Negative buffer the polygons by -X distance. This essentially smooths and combines smaller polygons. Hopefully, this approach will produce more linear, connected rows
  3. Select all polygons smaller than X area and delete them.

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.