YSBAT     Supervised Classification    02-1-07

- Explain the supervised classification process in 4 easy steps

- Explain why even relatively "pure" signatures have different distribution widths, and provide examples

- Describe the logic behind the minimum distance to means classifier, clarifying the difference between raw and standard deviation distances

- Using words and graphs, explain why the "standard deviation distance to means" is expected to do better than the "raw distance to means" when signatures range in distribution widths

- Using words and graphs, explain why the "standard deviation distance to means" is expected to do worse than the "raw distance to means" when some signatures are too wide

- Describe the logic behind the maximum likelihood classifier

- Describe the logic behind the parallel piped classifier

- Explain the difference between parametric and non-parametric classifiers

- Using words and signature statistics, frequency histograms, scattergrams, or "distance histograms" explain the three common problems in spectral signature development, including uncharacteristic training sites, more than one cover per signature, and non-represented covers