Classification models
Classification assigns one categorical class to each output pixel. The result is a class map, usually rendered with a different color for each class.
Requirements
- A finalized dataset version with labeled examples for the classes.
- Representative training and validation data for each class.
- A processing level and wavelength set appropriate for the imagery used later.
Labeled pixels do not need to be described as chemically pure. They should instead be accurate examples of the class definition you intend the model to learn.
Evaluation
Review overall and per-class precision, recall, F1, and support. Use the confusion matrix to find classes the model confuses. Accuracy alone can be misleading when classes are imbalanced.