I have been experimenting with interactive supervised image classification on a set of 4 band ortho images. I have received some good advice here but continue to struggle with some issues and I would like to start from scratch, as it were.
My imagery is a set of 150 orthos. They are four band images with a resolution of .25 cm per pixel.
- Is it best to pull these images together into a mosaic before supervised image classification?
- Are there optimal settings for the mosaic that prepare it for the classification process?
- What about exporting from the mosaic to a new raster with lower resolution - is that a better approach?
I have field data to support drawing a nice set of training polygons.
- How do you determine the optimal size for a training polygon?
- How do you determine the optimal number of training polygons for each class?
In my experiments so far, the main struggles I have had are with correctly classifying areas in shadow and areas that are generally brighter than the average pixel (e.g. south facing slopes). Any general advice on how to smooth this dynamic range out to be more consistently mapped?
Are there any new tools available in ArcGIS Pro 2 that may help me with this process? I have been using 10.5 for this project.
Any advice appreciated.