A comparison of procedures for classifying remotely-sensed data using simulated data sets

In this paper we present a systematic study of the efficacy of a number of different land-cover classification procedures by using simulated spectral responses. By simulation we have compared a minimum distance classifier, a maximum likelihood classifier and a penalized maximum likelihood classifier on responses from four different land-cover maps. The responses have been modelled in two ways, the second incorporating horizontal striping as a model of detector effects.