Category Decomposition based on Maximum Likelihood Estimation

Abstract

To make the accurate retrieval of the statistically meaningful proportion of each category among MIXEL's, a category decomposition method based on maximum likelihood estimation is proposed. In this method, the observed multispectral vector is considered as a probability variable along with the approximation that each pure pixel value is characterized by normal distribution. The results show that this method can retrieve more accurate proportion than previous method. And 2 indices that estimate the degree of error are proposed. The first one is always available to show the degree of estimation error in each category, and the second one is not always available to estimate the degree of estimation error of each MIXEL.