Model types

Choose a model type based on the value you want for each output pixel.

Model type	Output	Dataset requirement	Typical use
Classification	One categorical class	Labeled examples for the classes	Material or land-cover maps
Unmixing	One continuous response layer per class	At least two classes	Mixed-pixel and material-response analysis
Target Detection	A score for one target class	One target and one or more non-target classes	Locating a selected material
Regression	One continuous layer per numeric attribute	Numeric common label attributes and no reference spectra	Concentration or other quantitative estimates

Availability can depend on your account's model permissions.

Before choosing

Ask what the ground truth represents:

• Use Classification when every labeled pixel has one categorical answer.
• Use Unmixing when more than one material can contribute to a pixel and separate continuous responses are useful.
• Use Target Detection when only one class is the target and everything else supplies the non-target examples.
• Use Regression when labels carry numeric measurements rather than class membership.

Outputs are model estimates. Validate them against representative held-out data before using them as calibrated probabilities, abundances, or physical measurements.