An important application of the images taken by Landsat is the formation and updating of resource maps. This usually involves the assignment or classification of the individual picture elements (pixels) of the image to one of the ground classes of interest according to the values of the spectral intensities. A number of measures have been proposed for the overall performance of the classifier; these include the probability of error or misclassification. In certain cases, it is of interest to associate a measure of confidence or reliability in the classification of the individual pixels. For example, this measure could be used to compare classifications made from different images on an indivudual pixel level and so increase the overall accuracy of classification. To do this a distribution-free measure of reliability is required. A measure that is based on the comparison with the neighbourhood of the mode spectral vector of the class is proposed. Results from simulated and multispectral Landsat data are presented.
A measure of reliability for classification of Earth satellite data
Ein Mass fuer die Zuverlaessigkeit der Klassifizierung der Daten von Erdsatelliten
IEEE Transactions on Systems, Man and Cybernetics ; SMC-11 , 4 ; 312-318
1981
7 Seiten, 13 Quellen
Article (Journal)
English
MEASURE OF SATELLITE DISPERSION
AIAA | 1963
|Open data policies and satellite Earth observation
Elsevier | 2015
|