A new supervised classification method for quantitative analysis of remotely-sensed multi-spectral data

A new supervised classification method is developed for quantitative analysis of remotely-sensed multi-spectral data. It is based on the comparisons of the probability density function of the mixture of three normal distributions for a pixel and the probability density functions of the mixture of three normal distributions for spectral classes. The comparisons are made according to the distances between them. The discriminant function, which takes values on the interval [0, 2], is defined as Hellinger distance. The decision rule is established according to the values of Hellinger distances. The values of the discriminant functions give extra information including spectral similarity and difference percentages in the comparisons. This clarifies the classification results and could help researchers interpret better the classification results of remotely-sensed multi-spectral data. © 1998, Taylor & Francis Group, LLC. All rights reserved