Unidimensional scaling classifier and its application to remotely sensed data

Unidimensional Scaling (UDS) is to arrange n objects on the real line so that there inter-points distances are as close as possible to their observed distances. In this paper, we apply, this new method to remotely-sensed data, and then improve on this method according to the characteristics of remotely, sensed data. For validating this method, we make use of simulation data and remotely sensed data and compare the classification result with the method of K-Means. The result of comparison makes clear that UDS is succinctness and more understandable, and it can not only obtain the classification result of higher accuracy, but also be of the characteristics that it is not necessary, the prior information of class numbers and independent on the structure of data classified. At the same time, it hits advantage over the nonlimitness of high dimension of feature space.