Drone imagery segmentation over water

02-26-2021 02:04 PM
Labels (2)
New Contributor

First time posting...I'm trying to segment very high res (1.1cm/px) drone imagery over land and water, and can't for the life of me get a segmented result that makes any sense. The preview looks great (and I'm sure to always use 1:1 source resolution during processing) and has the desired boundaries. However once run, the segmented output looks terrible. I've tried changing the detail and segment size settings countless ways. Any clue what's going on here? Please see photos below. Many thanks for ANY guidance!

0 Kudos
1 Reply
New Contributor III

I'm having a very similar issue with some object-based classification in an urban setting. segmented raster looks great but final classified raster is terrible. It's always a game of adjustments.

Few questions:

- How many bands does your raster have? Is it just RGB, or multispectral?

- What exactly is the focus of your classification?

- How have you defined your training samples?

These may help you refine the segmentation and classification. If you have more than 3 bands, you can combine them in various ways in the Raster Symbology pane to highlight your features of interest, or at least get the water to mostly classify as one color. This choice of band combination will be driven by your classification focus (i.e. what features are most important to you in classifying/what is your classification supposed to achieve).

I see that your preview_classified image shows the water classified as multiple classes, when I'm sure you don't want that.

0 Kudos