DER (Dynamic Evidential Reasoning), applied to the classification of hyperspectral images

This paper describes a novel method for classification based on the evidential reasoning theory and the implementation presented by Peddle and Franklin. DER (Dynamic Evidential Reasoning) introduces some variations. It allows the incorporation of new evidence for the classifier in order to increment its accuracy, and it also defines a different decision rule. The inclusion of new evidence is a learning process where the precision of the classifier is analyzed using the Khat indicator. For this process, a set of samples belonging to a given, and a priori known, class is needed. This process changes the discriminate functions. The decision rule introduces two stages of decision. One stage is called "reject decision", where the maximum support value is analyzed, and the object to be classified is assigned to the unknown class if this support value is not "enough" (different approaches for the meaning of "enough" were studied). The other stage is called ambiguity decision, and the similarity between the maximum support and other supports of the rest of the classes are analyzed here. In this paper, an application of this method is presented, particularly, for the classification of different crops (during the growing season) in the area of Nebraska, using hyperspectral images. Some results are presented