Is there a method of targeting only specific areas for classification in eCognition? Say, maybe using pixel coordinate constraints?
To generally subset your data at the initial step, when you load in your data, in the top right corner, there will be the option to select a spatial subset.
Through that, you can either choose it based on pixel coordinates, projected coordinates, or simply draw a box.
If you wish to work within a smaller area of an existing classification, you can assign the objects to a new class, or even transfer them to a different level using a simple Assign Class approach.
For this, the use of temporary classification classes are recommended.