Resolving land use/land cover misclassification

Question

I'm relatively new to RS, I've been using Landsat Collection 2 Tier 2 data for my land use maps in ArcGIS Desktop. The problem is that my AOT has scarce rural built up area, due to which other barren areas (particularly close to glacier regions) keep getting misclassified as built up. I'm using the 6-4-2 band combination, and planetscope imagery (high res imagery for ancillary data)

How do I correct this? I have used Interactive classification feature a lot to correct the classification but a lot of areas still end up getting misclassified. I have used a hell lot of training areas in the regions that have misclassification but it's still not enough!

The main important land class is built up (only 0.3-2%) of the total area. Should I try object oriented classification? If yes, how do I do it in ArcGIS Desktop?

Answer 1

I have some experience in land classification in both QGIS SCP plugin and in ERDAS Imagine, never ArcGIS Suite. When I've run into issues similar to this one (such as deciduous vs. coniferous, or golf course vs. crops), the first thing I try is to experiment with the band configuration to see if there is a more desirable distinction between the classes of concern (which it seems is already within your purview), my next step would be to take the most accurate example I've had yet and bring it into QGIS with the imagery to manually edit the classification results.

So my answer to you is to take the best classification, maybe convert it into vectors (if it isn't already), and correct it manually as best you can given the circumstances/needs at-hand. Hopefully, the best classification output isn't too far off, and the task isn't too gargantuan.

Best of luck!