Classification of multispectral data by the use of a maximum likelihood classifier is dependent upon knowing in advance a set of prior probabilities. Therefore, the selection of an optimal set of prior probabilities is critical to the estimation of proportions for each class. In the proposed procedure, a function is minimized to yield a set of optimal prior probabilities for a specific data set. Classification results using optimal, actual, and default (equal prior probabilities for each class) values are compared.
Date of this Version