Extract Poly from classified image

[AtenaH__Talab]

Hi there,

I have an aerial imagery from a crop field.

I classified the image in order to extract crop's plots. (Discriminate plots from soil and shadow). I attach a snapshot of the image to give you an idea of what I mean by plots.

The Iso Cluster classification works really well. Yet I need to have each plot individually, I mean I need to extract these plots as separate polygons. The classification only gives me 3 classes (crop plots, shadow, and soil). Is there anyway that I could extract individual plots from the image?

Thank you so much

[IanMurray]

Once you have a classified image, you can convert it to polygon with the raster to polygon tool.  Each crop plot should be converted into its own polygon, as long as the cells of crop plots are not touching.  You would probably need to remove all the polygons for the other values(shadow and soils), and manually cut up polygons where the crop plots connect to each other if you need to, but it should give you each plot as its own polygon in a shapefile or gdb feature class.  If you need each to be its own shapefile or FC, that would take a little more work.

[DanPatterson_Retired]

Your image still looks like it could use some filtering...in any event, Region Group—Help | ArcGIS for Desktop would be the spatial analyst tool to group areas of common connectivity and value producing a grid of zones where each discrete grouping is given a unique value.

[DanPatterson_Retired]

Expand—Help | ArcGIS for Desktop  the raster by 1 cell

Shrink—Help | ArcGIS for Desktop  the raster by 1 celll

did it clean things up and not join the larger areas together?  try again  or try one of the other tools here

An overview of the Generalization toolset—Help | ArcGIS for Desktop

where RegionGroup, and Nibble are found along with MajorityFilter etc are found

Once you are done cleaning up the raster, then convert to vector, where you may have to fiddle with stuff there

Takes no time and is faster than trying to clean up vector files. 

There are also other useful methodologies but then require using NumPy arrays which at this stage is not a real option.

[DarrenWiens2]

Along the lines you (and the rest of this thread) have outlined, this help page goes through the steps to clean up your image. One of my most visited pages, I'm sure.

[DanPatterson_Retired]

Tnanks Darren.  Yes, it at the bottom of the tree in my​ link, but "explicit is better than implicit"

[Ross_PaulMartin]

You might want to look at feature analyst from Texatron. This is exactly why it was made. Dave Opitz solved this problem about 10 years ago.