Supervised Training and Classification
Selection of Training Areas
DN’s of training fields plotted on a “scatter” diagram in two-dimensional feature space Band 2 Band 1 from Lillesand & Kiefer
Classification Algorithms/Decision Rules Non-parametric decision rule – independent of the properties/statistics of the data – does not assume normal distribution – Example: Parallelepiped Parametric decision rule – assumes normal distribution – defined by the signature mean vector and covariance matrix – Examples: Minimum distance, Mahalanobis distance, Maximum likelihood
Parallelepiped “In the parallelepiped decision rule, the data file values of the candidate pixel are compared to upper and lower limits” the minimum and maximum data file values of each band in the signature - the mean of each band, plus and minus a number of standard deviations
Parallelepiped Fastest of all classifiers Problem of “overlapping bounds” Problem of “corner pixels”
Minimum Distance “Calculates the spectral distance between the measurement vector for the candidate pixel and the mean vector for each signature.” Advantages No unclassified pixels Disadvantages Does not incorporate variation Pixels that should be unclassified become classified
Maximum Likelihood For each pixel to be classified: The probability of classification is calculated for each class The pixel is classified as the class with the largest probability The slowest of classifiers discussed Theoretically the best classification
Maximum Likelihood Degrees of Probability
Maximum Likelihood
Maximum Likelihood Water Forest Urban 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 Band 4 - DN