novel confidence estimation method for neural networksin multispectral image classification

摘要

The Digital Earth concept has attracted much attention recently and thisapproach uses a variety of earth observation data from the global to the localscale. Imaging techniques have made much progress technically and the methodsused for automatic extraction of geo-ralated information are of importance inDigital Earth science. One of these methods, artificial neural networks (ANN)techniques, have been effectively used in classification of remotely sensed images.Generally image classification with ANN has been producing higher or equalmapping accuracies than parametric methods. Comparative studies have, in fact,shown that there is no discernible difference in classification accuracies betweenneural and conventional statistical approaches. Only well designed and trainedneural networks can present a better performance than the standard statisticalapproaches. There are, as yet, no widely recognised standard methods toimplement an optimum network. From this point of view it might be beneficialto quantify ANN's reliability in classification problems. To measure the reliabilityof the neural network might be a way of developing to determine suitable networkstructures. To date, the problem of confidence estimation of ANN has not beenstudied in remote sensing studies. A statistical method for quantifying thereliability of a neural network that can be used in image classification isinvestigated in this paper. For this purpose the method is to be based on abinomial experimentation concept to establish confidence intervals. This novelmethod can also be used for the selection of an appropriate network structure forthe classification of multispectral imagery. Although the main focus of theresearch is to estimate confidence in ANN, the approach might also be applicableand relevant to Digital Earth technologies.