Combining parametric and nonparametric classifiers for an unsupervised updating of land-cover maps

In this paper, the problem of unsupervised retraining of supervised classifiers for the analysis of multitemporal remote-sensing images is considered. In particular, two techniques are proposed for the unsupervised updating of the parameters of the maximum-likelihood and the radial basis function neural-network classifiers, on the basis of the distribution of a new image to be classified. Given the complexity inherent with the task of unsupervised retraining, the resulting classifiers are intrinsically less reliable and accurate than the corresponding supervised approaches, especially for complex data sets. In order to overcome this drawback, we propose to use methodologies for the combination of different classifiers to increase the accuracy and the reliability of unsupervised retraining classifiers. This allows one to obtain in an unsupervised way classification performances close to the ones of supervised approaches. © Springer-Verlag Berlin Heidelberg 2000.